live reporting commit requests, resume prints batching

- Persist loop-active state for widget re-instantiation after session reload
- Add YAML task file parsing and update support via yaml library
- Auto-update PRD source file checkboxes on task status changes
- Add batchRender callback for real-time parallel widget animation
- Normalize tabs-to-spaces indentation across source files
- Use padStart(2, '0') for ID formatting instead of hardcoded prefix
- Enable parallel execution for single-task DAG batches

.gitignore

@@ -1,4 +1,3 @@
 node_modules
-dist
 .pi-lens
 package-lock.json

AGENTS.md

@@ -4,18 +4,17 @@
 
 A Pi coding agent extension that registers the `/ralpi` slash command. Not a standalone app — it runs inside Pi's extension host.
 
-## Build
+## Type checking
 
 ```
-npm run build    # tsc → dist/
-npm run watch    # tsc --watch
+npm run typecheck    # tsc --noEmit
 ```
 
-No bundler, no linter, no test framework. Plain `tsc` with strict mode.
+No build step needed — Pi loads extensions via [jiti](https://github.com/unjs/jiti), which compiles TypeScript at runtime. `index.ts` is the entry point directly.
 
 ## Entry point
 
-`index.ts` at repo root (not `src/`). Exports a default function receiving `ExtensionAPI`. The `tsconfig.json` sets `rootDir: "./"` so `index.ts` compiles to `dist/index.js`.
+`index.ts` at repo root (not `src/`). Exports a default function receiving `ExtensionAPI`.
 
 ## External dependencies
 
@@ -27,7 +26,7 @@ The only real npm dependency is `yaml` (^2.4.0).
 
 ## Source structure
 
-- `index.ts` — extension entry, command routing, UI registration
+- `index.ts` — extension entry, command routing, UI registration, reload detection
 - `src/` — all logic modules:
   - `parser.ts` — task file parsing (Fio, checkbox, YAML formats)
   - `dag.ts` — Kahn's algorithm dependency resolution, batch planning
@@ -38,12 +37,13 @@ The only real npm dependency is `yaml` (^2.4.0).
   - `utils.ts` — config loading, progress discovery, `runAgentSession()`
   - `types.ts` — all interfaces and `DEFAULT_CONFIG`
   - `widget-batcher.ts` — debounced widget updates for parallel tasks
+  - `constants.ts` — static constants
 - `skills/ralpi-use.md` — Pi skill definition for task execution
-- `tasks/` — example ralpi task files (self-modification history)
+- `prompts/task-manager.md` — Pi prompt for task planning
 
 ## Runtime state
 
-All runtime state lives in `.ralpi/` (gitignored):
+All runtime state lives in `.ralpi/` in the **project directory** (not this extension directory):
 - `.ralpi/progress.json` — execution progress, supports multiple PRDs
 - `.ralpi/reflections/` — per-task reflection JSON files
 - `.ralpi/prompts/` — generated prompts (timestamped, for debugging)
@@ -55,7 +55,7 @@ Task IDs are zero-padded strings (`"01"`, `"02"`, etc.). The parser prepends `0`
 
 ## Command routing
 
-`/ralpi` with no args → plan. First token looks like a path (`@path`, `./path`, `.md`, etc.) → run. Otherwise dispatches to subcommand (`run`, `plan`, `status`, `resume`, `next`, `reset`).
+`/ralpi` with no args → plan. First token looks like a path (`@path`, `./path`, `.md`, etc.) → run. Otherwise dispatches to subcommand (`run`, `plan`, `resume`, `reset`).
 
 ## Config
 

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Michael Freno
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -1,78 +1,36 @@
 # Ralpi
 
-Execute tasks from task files using DAG-based dependency resolution with persistent progress tracking.
+Execute tasks from task files until done using DAG-based dependency resolution with persistent progress tracking.
+
+```bash
+pi install npm:@mikefreno/ralpi
+```
 
 ## Features
 
-- **DAG-based execution**: Tasks are ordered by dependencies using Kahn's algorithm
 - **Parallel batching**: Independent tasks in each batch can run concurrently
 - **Persistent progress**: Execution state saved to `.ralpi/progress.json`
 - **Reflection system**: Each task produces a reflection for downstream tasks
 - **Retry with backoff**: Failed tasks retry with exponential backoff
-- **Multiple formats**: Supports Fio README, simple checkboxes, and YAML
-- **Chat progress**: Real-time progress messages in Pi chat via `pi.sendMessage`
+- **Multiple formats**: Supports simple checkboxes, and YAML
 - **Tool usage tracking**: Detects and reports tool usage (read, write, edit, bash) from task execution
-- **Git commit capture**: Captures git commit messages and generates summaries per task
 - **Configurable timeouts**: Task-level timeouts via meta blocks, with global fallback
 - **Session saving**: Saves full task output for expandable session review
 - **Resume auto-discovery**: Automatically finds and resumes interrupted execution
-- **Custom message renderer**: Compact UI labels with expandable details in Pi TUI
 
 ## Usage
 
 ```
-/ralpi plan [task-file]   # Show execution plan
-/ralpi run [task-file]    # Execute all tasks
-/ralpi status [task-file] # Show current progress
-/ralpi resume [task-file] # Resume paused execution
-/ralpi next [task-file]   # Execute next batch only
-/ralpi reset [task-file]  # Reset all progress
+/ralpi [task-file]  # Execute all tasks
+/ralpi plan         # Alias to /task-manager to plan new tasks
+/ralpi resume       # Resume paused execution
+/ralpi reset        # Reset progress and .ralpi directory - does not modify PRD
 ```
 
-## Task File Formats
-
-### Fio README Format
-
-```markdown
-# Project Title
+### Highly recommended to use the task-manager prompt for prd construction, it's output pairs perfectly - /task-manager or /ralpi plan
 
 ## Tasks
 
-- [ ] 01 — Setup project structure -> `tasks/01-setup.md`
-- [ ] 02 — Implement auth -> `tasks/02-auth.md`
-- [ ] 03 — Build API -> `tasks/03-api.md`
-
-## Dependencies
-
-1 -> 2,3
-2 -> 3
-```
-
-#### Supported Dependency Formats
-
-The parser supports two dependency declaration styles in the `## Dependencies` section:
-
-**Arrow Notation** (recommended):
-```
-1 -> 2,3,4
-5 -> 6
-```
-This means: "Task 1 must complete before tasks 2, 3, and 4 can start."
-
-**Natural Language**:
-```
-13 depends on 17, 18, 19, 20
-14 depends on 13, 15, 16
-```
-This means: "Task 13 depends on tasks 17, 18, 19, and 20."
-
-**Parallel Groups** (informational only):
-```
-1, 2, 3, 4 can be done in parallel
-5, 6, 7, 8 can be done in parallel
-```
-Note: These lines are ignored by the parser. Use explicit dependencies to control execution order.
-
 ### Simple Checkbox Format
 
 ```markdown
@@ -96,17 +54,47 @@ tasks:
     depends_on: ["01"]
 ```
 
-## Configuration
+## Task IDs
 
-Create `.ralpi/config.yaml`:
+Task IDs are zero-padded 2-digit strings (`01`, `02`, ...) with an optional
+single lowercase letter suffix for sub-tasks inserted between two numbered
+steps (e.g. `02b`, `02c`). The parser normalizes `2b` → `02b`.
 
-```yaml
-maxRetries: 3
-retryDelayMs: 5000
-timeoutMs: 1800000
-maxParallel: 3
-projectContext: "Additional context for all tasks"
 ```
+- [ ] 01 — Setup
+- [ ] 02 — Fix bugs
+- [ ] 02b — Sub-step of 02 (inserted after the fact)
+- [ ] 02c — Another sub-step of 02
+- [ ] 03 — Continue
+```
+
+Use lettered sub-tasks when you discover mid-stream that a step needs to be
+split. They let you preserve sibling numbering (`01`, `02`, `03`, ...) while
+adding granularity between two existing steps.
+
+## Dependencies
+
+### Arrow Notation (recommended)
+
+1 -> 2,3,4
+5 -> 6
+This means: "Task 1 must complete before tasks 2, 3, and 4 can start."
+
+### Natural Language
+
+13 depends on 17, 18, 19, 20
+14 depends on 13, 15, 16
+
+This means: "Task 13 depends on tasks 17, 18, 19, and 20."
+
+### Parallel Groups (informational only)
+
+1, 2, 3, 4 can be done in parallel
+5, 6, 7, 8 can be done in parallel
+
+Note: These lines are ignored by the parser. Use explicit dependencies to control execution order.
+
+## Configuration
 
 ### Task-Level Timeout
 
@@ -119,6 +107,33 @@ You can set a timeout for individual tasks using a meta block in the task file:
 
 Supported formats: `10m` (minutes), `600s` (seconds), `3600000` (milliseconds)
 
+### Config files
+
+| Scope | Path |
+|-------|------|
+| **Global** | `~/.pi/ralpi/config.yaml` |
+| **Project** | `./.ralpi/config.yaml` |
+
+```yaml
+execution:
+  maxParallel: 3          # ralpi-level concurrency only
+  models:                 # round-robin in <provider>/<model> format
+    - google/gemini-3.5-flash # 1st and 3rd task in parallel
+    - openai/gpt-5.5 # 2nd task in parallel
+prompts:
+  projectContext: "Additional context for all tasks"
+```
+
+> `execution.models` uses slot-aware round-robin: with 3 models and 2 concurrent
+> tasks, only the first two models are used. The third model is only touched when
+> a third concurrent task starts. Freed model slots are reused before new ones
+> are allocated.
+> **Automatic failover**: if a provider/API is unreachable (rate limit, 503, etc.),
+> the task automatically cycles to the next model in the list without counting it
+> as a task failure. Each model is tried once before the task is marked as failed.
+> **NOTE**: this is only used in parallel execution, in sequential mode the
+> parent pi session's model is used
+
 ## State Files
 
 - `.ralpi/progress.json` - Execution progress

bun.lock

@@ -0,0 +1,308 @@
+{
+  "lockfileVersion": 1,
+  "configVersion": 1,
+  "workspaces": {
+    "": {
+      "name": "@mikefreno/ralpi",
+      "dependencies": {
+        "yaml": "^2.4.0",
+      },
+      "devDependencies": {
+        "@types/node": "^20.0.0",
+        "bun-types": "^1.3.14",
+        "typescript": "^5.3.0",
+      },
+      "peerDependencies": {
+        "@earendil-works/pi-coding-agent": "*",
+        "@earendil-works/pi-tui": "*",
+      },
+    },
+  },
+  "packages": {
+    "@anthropic-ai/sdk": ["@anthropic-ai/sdk@0.91.1", "", { "dependencies": { "json-schema-to-ts": "^3.1.1" }, "peerDependencies": { "zod": "^3.25.0 || ^4.0.0" }, "optionalPeers": ["zod"], "bin": { "anthropic-ai-sdk": "bin/cli" } }, "sha512-LAmu761tSN9r66ixvmciswUj/ZC+1Q4iAfpedTfSVLeswRwnY3n2Nb6Tsk+cLPP28aLOPWeMgIuTuCcMC6W/iw=="],
+
+    "@aws-crypto/crc32": ["@aws-crypto/crc32@5.2.0", "", { "dependencies": { "@aws-crypto/util": "^5.2.0", "@aws-sdk/types": "^3.222.0", "tslib": "^2.6.2" } }, "sha512-nLbCWqQNgUiwwtFsen1AdzAtvuLRsQS8rYgMuxCrdKf9kOssamGLuPwyTY9wyYblNr9+1XM8v6zoDTPPSIeANg=="],
+
+    "@aws-crypto/sha256-browser": ["@aws-crypto/sha256-browser@5.2.0", "", { "dependencies": { "@aws-crypto/sha256-js": "^5.2.0", "@aws-crypto/supports-web-crypto": "^5.2.0", "@aws-crypto/util": "^5.2.0", "@aws-sdk/types": "^3.222.0", "@aws-sdk/util-locate-window": "^3.0.0", "@smithy/util-utf8": "^2.0.0", "tslib": "^2.6.2" } }, "sha512-AXfN/lGotSQwu6HNcEsIASo7kWXZ5HYWvfOmSNKDsEqC4OashTp8alTmaz+F7TC2L083SFv5RdB+qU3Vs1kZqw=="],
+
+    "@aws-crypto/sha256-js": ["@aws-crypto/sha256-js@5.2.0", "", { "dependencies": { "@aws-crypto/util": "^5.2.0", "@aws-sdk/types": "^3.222.0", "tslib": "^2.6.2" } }, "sha512-FFQQyu7edu4ufvIZ+OadFpHHOt+eSTBaYaki44c+akjg7qZg9oOQeLlk77F6tSYqjDAFClrHJk9tMf0HdVyOvA=="],
+
+    "@aws-crypto/supports-web-crypto": ["@aws-crypto/supports-web-crypto@5.2.0", "", { "dependencies": { "tslib": "^2.6.2" } }, "sha512-iAvUotm021kM33eCdNfwIN//F77/IADDSs58i+MDaOqFrVjZo9bAal0NK7HurRuWLLpF1iLX7gbWrjHjeo+YFg=="],
+
+    "@aws-crypto/util": ["@aws-crypto/util@5.2.0", "", { "dependencies": { "@aws-sdk/types": "^3.222.0", "@smithy/util-utf8": "^2.0.0", "tslib": "^2.6.2" } }, "sha512-4RkU9EsI6ZpBve5fseQlGNUWKMa1RLPQ1dnjnQoe07ldfIzcsGb5hC5W0Dm7u423KWzawlrpbjXBrXCEv9zazQ=="],
+
+    "@aws-sdk/client-bedrock-runtime": ["@aws-sdk/client-bedrock-runtime@3.1048.0", "", { "dependencies": { "@aws-crypto/sha256-browser": "5.2.0", "@aws-crypto/sha256-js": "5.2.0", "@aws-sdk/core": "^3.974.11", "@aws-sdk/credential-provider-node": "^3.972.42", "@aws-sdk/eventstream-handler-node": "^3.972.16", "@aws-sdk/middleware-eventstream": "^3.972.12", "@aws-sdk/middleware-websocket": "^3.972.19", "@aws-sdk/token-providers": "3.1048.0", "@aws-sdk/types": "^3.973.8", "@smithy/core": "^3.24.2", "@smithy/fetch-http-handler": "^5.4.2", "@smithy/node-http-handler": "^4.7.2", "@smithy/types": "^4.14.1", "tslib": "^2.6.2" } }, "sha512-u+NT61JZEkRFtpL0CAw1N1dwxnaLgwVXQl/zjJxTGgLyS/jTIdg2SdoEoCTHxgDyCnqa1HEi9QOoE9/pYRNpOQ=="],
+
+    "@aws-sdk/core": ["@aws-sdk/core@3.974.19", "", { "dependencies": { "@aws-sdk/types": "^3.973.12", "@aws-sdk/xml-builder": "^3.972.29", "@aws/lambda-invoke-store": "^0.2.2", "@smithy/core": "^3.24.6", "@smithy/signature-v4": "^5.4.6", "@smithy/types": "^4.14.3", "bowser": "^2.11.0", "tslib": "^2.6.2" } }, "sha512-SMNfLCU/41xxfFaC5Slwy8V/f1FRhakvyeeMeDeIxqNF0DzhDlXsXnJDELJYke1EtnJbfzfilW7tvulGfxMY6A=="],
+
+    "@aws-sdk/credential-provider-env": ["@aws-sdk/credential-provider-env@3.972.45", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-ZPsnLyrpDRmojKrBbJykASyLLVFkjyD+fWATeSuYgaqablijGOzxPxEKyrwUvNg+bgSQ7PkW2FTu65Xco19Gag=="],
+
+    "@aws-sdk/credential-provider-http": ["@aws-sdk/credential-provider-http@3.972.47", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/fetch-http-handler": "^5.4.6", "@smithy/node-http-handler": "^4.7.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-1XdgHDIPbARHuzZXM7ouzIbSUZFU9dTi9k+ryMhiZU4QCam4dvwOyUEFjEHNxAZehCYUIOmsSUZ2un6BIgUkWg=="],
+
+    "@aws-sdk/credential-provider-ini": ["@aws-sdk/credential-provider-ini@3.972.51", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/credential-provider-env": "^3.972.45", "@aws-sdk/credential-provider-http": "^3.972.47", "@aws-sdk/credential-provider-login": "^3.972.50", "@aws-sdk/credential-provider-process": "^3.972.45", "@aws-sdk/credential-provider-sso": "^3.972.50", "@aws-sdk/credential-provider-web-identity": "^3.972.50", "@aws-sdk/nested-clients": "^3.997.18", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/credential-provider-imds": "^4.3.7", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-f8sRTVyM+9BbzQKPlUP9dVVpgNEu65jFckNAAGzRfCrlaSi5AWUbCKEHIMcIYokv8pWblSKEqHkqKYZtwINnhw=="],
+
+    "@aws-sdk/credential-provider-login": ["@aws-sdk/credential-provider-login@3.972.50", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/nested-clients": "^3.997.18", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-NHHsKoMhw6UylSU0XDnDc87+IQW8tRBTIe6vnOX12GSIlBDtoce6bSzONleIglCyu8d3H9bmTSfk+sIN5yh3WA=="],
+
+    "@aws-sdk/credential-provider-node": ["@aws-sdk/credential-provider-node@3.972.53", "", { "dependencies": { "@aws-sdk/credential-provider-env": "^3.972.45", "@aws-sdk/credential-provider-http": "^3.972.47", "@aws-sdk/credential-provider-ini": "^3.972.51", "@aws-sdk/credential-provider-process": "^3.972.45", "@aws-sdk/credential-provider-sso": "^3.972.50", "@aws-sdk/credential-provider-web-identity": "^3.972.50", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/credential-provider-imds": "^4.3.7", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-z/JJ8Qvf2GiTn4bw+x8k7wQjxmPpNsiwZ7ls/h1cZHikrSpS0+65lB+lafnXZlxv1lqH4k6rQwh+2UsycC662g=="],
+
+    "@aws-sdk/credential-provider-process": ["@aws-sdk/credential-provider-process@3.972.45", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-QMJXjTGLmHE4Ie03T5H4hHOLfcvMc9DaODO6b5dgte3S8ECf5bBuHUJW4cQREcYZyRkOU8iymqtiBxqF4icxZg=="],
+
+    "@aws-sdk/credential-provider-sso": ["@aws-sdk/credential-provider-sso@3.972.50", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/nested-clients": "^3.997.18", "@aws-sdk/token-providers": "3.1064.0", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-pQ9ww4G53gwHlon1NMz25JhaBo13E9Jv+VVgjh39C/yzvby+xhSnEOb+VDYShKNCh1TbttMF/5CFCHkZrIqOcA=="],
+
+    "@aws-sdk/credential-provider-web-identity": ["@aws-sdk/credential-provider-web-identity@3.972.50", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/nested-clients": "^3.997.18", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-9DbaPaT2aMbz18wtSpq9HVBErjBQwxykqTFgG6n8Bn05GN68mITz+G1869ekYx0mVT/BDjETj5czz/3cPgLwxA=="],
+
+    "@aws-sdk/eventstream-handler-node": ["@aws-sdk/eventstream-handler-node@3.972.21", "", { "dependencies": { "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-mVC0hOmwGJmNFezZ+wM8Sqfap/LjsMavEf2Evl0YWrLAcrdZOEdjnY8nRvgakVViWJSGm2eJxLuPVHGdeV06kA=="],
+
+    "@aws-sdk/middleware-eventstream": ["@aws-sdk/middleware-eventstream@3.972.17", "", { "dependencies": { "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-tdbnXbw73ww62ABWP0G0Z/euvFowEEvAoi/zG4NaZo7HJFpfGho/Z65HyVzkJLT1cMsUregr4pTyxljlarT0wA=="],
+
+    "@aws-sdk/middleware-websocket": ["@aws-sdk/middleware-websocket@3.972.27", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/fetch-http-handler": "^5.4.6", "@smithy/signature-v4": "^5.4.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-V/IgUogQm/NSGlNglDCkREirQXgjyrrq64vPt5qcRTGhEJJwPcUB3RyYE6iZ63WNGp4Ezc+JtVRA4QlSbhDvVQ=="],
+
+    "@aws-sdk/nested-clients": ["@aws-sdk/nested-clients@3.997.18", "", { "dependencies": { "@aws-crypto/sha256-browser": "5.2.0", "@aws-crypto/sha256-js": "5.2.0", "@aws-sdk/core": "^3.974.19", "@aws-sdk/signature-v4-multi-region": "^3.996.33", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/fetch-http-handler": "^5.4.6", "@smithy/node-http-handler": "^4.7.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-xBWrodBvW5SHCZV11UZUJG0pSHkLCEREIBoNbff1C1sacOUCmxJnTCPE80sCGLCtqgXg98I2MQJe2z28tcZSsw=="],
+
+    "@aws-sdk/signature-v4-multi-region": ["@aws-sdk/signature-v4-multi-region@3.996.33", "", { "dependencies": { "@aws-sdk/types": "^3.973.12", "@smithy/signature-v4": "^5.4.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-Hn0RThJEbyOZWV2PV9Z4YD3nitGPxybmyU17dSe9b61WOBcKnqS0WTtM3c1zyZq9WnGiyrfi/i+UBPUk7cM8Ug=="],
+
+    "@aws-sdk/token-providers": ["@aws-sdk/token-providers@3.1048.0", "", { "dependencies": { "@aws-sdk/core": "^3.974.11", "@aws-sdk/nested-clients": "^3.997.9", "@aws-sdk/types": "^3.973.8", "@smithy/core": "^3.24.2", "@smithy/types": "^4.14.1", "tslib": "^2.6.2" } }, "sha512-k0y/GcuesuSfWyUM0WamrGyeZmltRYaPbHO82UDA6mZ/doB+FOHKutikPAtSXMn/hDz970cF+iRuuiYO9VEbAA=="],
+
+    "@aws-sdk/types": ["@aws-sdk/types@3.973.12", "", { "dependencies": { "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-43ajd1NF0RMgX5k0hxCNUyEdrtFUsb2aHT2QvpktSC/2Eyb2Jr/JPVqdp0XIoaHWikZJq5tNWSLO6kB5q2eMCA=="],
+
+    "@aws-sdk/util-locate-window": ["@aws-sdk/util-locate-window@3.965.7", "", { "dependencies": { "tslib": "^2.6.2" } }, "sha512-M0D6oIpohdNHjc7udzTHEQyot0+0iuA36jc2I9Hps+f/GtKi2HO/pyijQnCnNcwZqLB5+rtn81z3eZK/GyjAmA=="],
+
+    "@aws-sdk/xml-builder": ["@aws-sdk/xml-builder@3.972.29", "", { "dependencies": { "@smithy/types": "^4.14.3", "fast-xml-parser": "5.7.3", "tslib": "^2.6.2" } }, "sha512-fk0niuGFxfi8yIJuMVM4mhwObkiQSuwZFj3tAPrLVx64Pk3BkrEIpqjzHKY4hKoEBUD6Jg/S74Zj9jy+5F3DnQ=="],
+
+    "@aws/lambda-invoke-store": ["@aws/lambda-invoke-store@0.2.4", "", {}, "sha512-iY8yvjE0y651BixKNPgmv1WrQc+GZ142sb0z4gYnChDDY2YqI4P/jsSopBWrKfAt7LOJAkOXt7rC/hms+WclQQ=="],
+
+    "@babel/runtime": ["@babel/runtime@7.29.7", "", {}, "sha512-Nq8OhGWiZIZGV6hLHoyAKLLcJihP/xFeBMGJoUrxTX2psI8dCifzLhZISFb+VWS3wFMRDmCGw5R+dOySCqPLhw=="],
+
+    "@earendil-works/pi-agent-core": ["@earendil-works/pi-agent-core@0.79.0", "", { "dependencies": { "@earendil-works/pi-ai": "^0.79.0", "ignore": "7.0.5", "typebox": "1.1.38", "yaml": "2.9.0" } }, "sha512-jQOtYjRGZ7+XC/olw9euLd2V03vkAPO8u0sSnQoLbyOQZz66dEBZrklTESk34Sf3AaeBSua28wjZR48ch1aXJQ=="],
+
+    "@earendil-works/pi-ai": ["@earendil-works/pi-ai@0.79.0", "", { "dependencies": { "@anthropic-ai/sdk": "0.91.1", "@aws-sdk/client-bedrock-runtime": "3.1048.0", "@google/genai": "1.52.0", "@mistralai/mistralai": "2.2.1", "@smithy/node-http-handler": "4.7.3", "http-proxy-agent": "7.0.2", "https-proxy-agent": "7.0.6", "openai": "6.26.0", "partial-json": "0.1.7", "typebox": "1.1.38" }, "bin": { "pi-ai": "dist/cli.js" } }, "sha512-D/2aDoe9vcCbqAztALQcKkdqXGuaQcqAzLm8LfUhNaorwoIHkwnaAuDVlo+OkF5clpEwS8Z1bk2o8NiSrwEdsA=="],
+
+    "@earendil-works/pi-coding-agent": ["@earendil-works/pi-coding-agent@0.79.0", "", { "dependencies": { "@earendil-works/pi-agent-core": "^0.79.0", "@earendil-works/pi-ai": "^0.79.0", "@earendil-works/pi-tui": "^0.79.0", "@silvia-odwyer/photon-node": "0.3.4", "chalk": "5.6.2", "cross-spawn": "7.0.6", "diff": "8.0.4", "glob": "13.0.6", "highlight.js": "10.7.3", "hosted-git-info": "9.0.3", "ignore": "7.0.5", "jiti": "2.7.0", "minimatch": "10.2.5", "proper-lockfile": "4.1.2", "typebox": "1.1.38", "undici": "8.3.0", "yaml": "2.9.0" }, "optionalDependencies": { "@mariozechner/clipboard": "0.3.9" }, "bin": { "pi": "dist/cli.js" } }, "sha512-pZoXk65vFR3dAzzmPNWEX61aHnT6+BaVhTyFDQAs1DyumaMeWpvzRV9ZrGxqlbVLwhrq+0LnXbaqDAFkhe2+MQ=="],
+
+    "@earendil-works/pi-tui": ["@earendil-works/pi-tui@0.79.0", "", { "dependencies": { "get-east-asian-width": "1.6.0", "marked": "15.0.12" } }, "sha512-qAQWMruW7YKbk2hPcTD4INtXfvIySXifbPQ+mFY5j3J8yf2tfElkh+gGPuBvgPKPT0z9WiAkd7iySCuQq0txuQ=="],
+
+    "@google/genai": ["@google/genai@1.52.0", "", { "dependencies": { "google-auth-library": "^10.3.0", "p-retry": "^4.6.2", "protobufjs": "^7.5.4", "ws": "^8.18.0" }, "peerDependencies": { "@modelcontextprotocol/sdk": "^1.25.2" }, "optionalPeers": ["@modelcontextprotocol/sdk"] }, "sha512-gwSvbpiN/17O9TbsqSsE/OzZcpv5Fo4RQjdngGgogtuB9RsyJ8ZHhX5KjHj1bp5N9snN2eK8LDGXSaWW2hof8Q=="],
+
+    "@mariozechner/clipboard": ["@mariozechner/clipboard@0.3.9", "", { "optionalDependencies": { "@mariozechner/clipboard-darwin-arm64": "0.3.9", "@mariozechner/clipboard-darwin-universal": "0.3.9", "@mariozechner/clipboard-darwin-x64": "0.3.9", "@mariozechner/clipboard-linux-arm64-gnu": "0.3.9", "@mariozechner/clipboard-linux-arm64-musl": "0.3.9", "@mariozechner/clipboard-linux-riscv64-gnu": "0.3.9", "@mariozechner/clipboard-linux-x64-gnu": "0.3.9", "@mariozechner/clipboard-linux-x64-musl": "0.3.9", "@mariozechner/clipboard-win32-arm64-msvc": "0.3.9", "@mariozechner/clipboard-win32-x64-msvc": "0.3.9" } }, "sha512-ABnA53mdfkGZwOFUdZNv2S0CWGO/EIuPj8Vv9xmBFmSYg/qFc7ihO6q5FcQjvoE67kZpWkEc4AhD6B/os04yuA=="],
+
+    "@mariozechner/clipboard-darwin-arm64": ["@mariozechner/clipboard-darwin-arm64@0.3.9", "", { "os": "darwin", "cpu": "arm64" }, "sha512-BfgV7vCEWZwJwZJw03r6bP5+tf0iI/ANuQYCxi9RNn7FrWB3yzGuMKCrNLRl6V761vXRdL8+OqZ0wd4TqlsNOQ=="],
+
+    "@mariozechner/clipboard-darwin-universal": ["@mariozechner/clipboard-darwin-universal@0.3.9", "", { "os": "darwin" }, "sha512-BGGR4iA9Z2shAjI65eI5xtyb3LYNlDW9X3gxKxDbqtbnREohsrqznov6zpKoIrsRWpzlYVEdKphS7ksJ0/ndSQ=="],
+
+    "@mariozechner/clipboard-darwin-x64": ["@mariozechner/clipboard-darwin-x64@0.3.9", "", { "os": "darwin", "cpu": "x64" }, "sha512-4kURmCbS6nt8uYhtmWpUcJWyPHfmAr5dTpXD1nO3pIfa+TSQ9DbrGOYCKH+aEFW47XhQ4Vp8ZTszie+wfFvDKg=="],
+
+    "@mariozechner/clipboard-linux-arm64-gnu": ["@mariozechner/clipboard-linux-arm64-gnu@0.3.9", "", { "os": "linux", "cpu": "arm64" }, "sha512-g59OkUGP2DDfCOIKypHeYgv2M55u/cKvXa5dSxFbEJ34XvIQMdcVmpKCkGUro3ZgefXiGVdwguvTMQGpHWzIXw=="],
+
+    "@mariozechner/clipboard-linux-arm64-musl": ["@mariozechner/clipboard-linux-arm64-musl@0.3.9", "", { "os": "linux", "cpu": "arm64" }, "sha512-AGuJdgKsmJdm4Pych7kv3sqe591ERRaAHW3xjLooiFzn8J+PxUyof++7YZrB5Y5tpnTO+K18Og3taj2NpluCRQ=="],
+
+    "@mariozechner/clipboard-linux-riscv64-gnu": ["@mariozechner/clipboard-linux-riscv64-gnu@0.3.9", "", { "os": "linux", "cpu": "none" }, "sha512-DXBEAiuMpk7dhS1a9NzNxVAFi1vaKoPu7rQNgY8LIDLGrK3lnIp3nT10DUum+PKVJoJppIP+NAA8IZe4DMNDPw=="],
+
+    "@mariozechner/clipboard-linux-x64-gnu": ["@mariozechner/clipboard-linux-x64-gnu@0.3.9", "", { "os": "linux", "cpu": "x64" }, "sha512-WORrMLd6EpElEME7JRKfSaY34nW1P5LbdgK5YNCS1ncG2LqmITsSMEJ8nh2mpvxb3TxqbOOKgY7k9eMJYlW9Mw=="],
+
+    "@mariozechner/clipboard-linux-x64-musl": ["@mariozechner/clipboard-linux-x64-musl@0.3.9", "", { "os": "linux", "cpu": "x64" }, "sha512-/DHn+1DrfL6oRaPPWXaOKvonFFrni666fxd+zFqiQEfvBH0tsHVWjq9iqBk0oDp0qaPA72lIMy5BptxISBEhZQ=="],
+
+    "@mariozechner/clipboard-win32-arm64-msvc": ["@mariozechner/clipboard-win32-arm64-msvc@0.3.9", "", { "os": "win32", "cpu": "arm64" }, "sha512-O5FHD3ErkMwMhNzAfu3ggy0ug4z7btZuoQgwwxlzPrwV2bxlD6WDpqBY4NCgICAgZdDKdp+loUEKVAVt8aYnhQ=="],
+
+    "@mariozechner/clipboard-win32-x64-msvc": ["@mariozechner/clipboard-win32-x64-msvc@0.3.9", "", { "os": "win32", "cpu": "x64" }, "sha512-ihQC3EufqEY81vhXBgVBtK4prL+wc62zJsSvxrgz7K1hsdt6OObz6v9p3Rn1OG3GJksTTKMJF0u/guMISHPhSA=="],
+
+    "@mistralai/mistralai": ["@mistralai/mistralai@2.2.1", "", { "dependencies": { "ws": "^8.18.0", "zod": "^3.25.0 || ^4.0.0", "zod-to-json-schema": "^3.25.0" } }, "sha512-uKU8CZmL2RzYKmplsU01hii4p3pe4HqJefpWNRWXm1Tcm0Sm4xXfwSLIy4k7ZCPlbETCGcp69E7hZs+WOJ5itQ=="],
+
+    "@nodable/entities": ["@nodable/entities@2.1.1", "", {}, "sha512-Pig3HxDIoMgjdEH8OCf/dkcTmLFjJRjWuq8jSnklu284/TKOPibSRERmOykiwmyXTtv61mP+44f3GMx0tLAyjg=="],
+
+    "@protobufjs/aspromise": ["@protobufjs/aspromise@1.1.2", "", {}, "sha512-j+gKExEuLmKwvz3OgROXtrJ2UG2x8Ch2YZUxahh+s1F2HZ+wAceUNLkvy6zKCPVRkU++ZWQrdxsUeQXmcg4uoQ=="],
+
+    "@protobufjs/base64": ["@protobufjs/base64@1.1.2", "", {}, "sha512-AZkcAA5vnN/v4PDqKyMR5lx7hZttPDgClv83E//FMNhR2TMcLUhfRUBHCmSl0oi9zMgDDqRUJkSxO3wm85+XLg=="],
+
+    "@protobufjs/codegen": ["@protobufjs/codegen@2.0.5", "", {}, "sha512-zgXFLzW3Ap33e6d0Wlj4MGIm6Ce8O89n/apUaGNB/jx+hw+ruWEp7EwGUshdLKVRCxZW12fp9r40E1mQrf/34g=="],
+
+    "@protobufjs/eventemitter": ["@protobufjs/eventemitter@1.1.1", "", {}, "sha512-vW1GmwMZNnL+gMRaovlh9yZX74kc+TTU3FObkkurpMaRtBfLP3ldjS9KQWlwZgraRE0+dheEEoAxdzcJQ8eXZg=="],
+
+    "@protobufjs/fetch": ["@protobufjs/fetch@1.1.1", "", { "dependencies": { "@protobufjs/aspromise": "^1.1.1" } }, "sha512-GpptLrs57adMSuHi3VNj0mAF8dwh36LMaYF6XyJ6JMWlVsc+t42tm1HSEDmOs3A8fC9yyeisgLhsTVQokOZ0zw=="],
+
+    "@protobufjs/float": ["@protobufjs/float@1.0.2", "", {}, "sha512-Ddb+kVXlXst9d+R9PfTIxh1EdNkgoRe5tOX6t01f1lYWOvJnSPDBlG241QLzcyPdoNTsblLUdujGSE4RzrTZGQ=="],
+
+    "@protobufjs/inquire": ["@protobufjs/inquire@1.1.2", "", {}, "sha512-pa0vFRuws4wkvaXKK1uXZMAwAX4/t8ANaJo45iw/oQHNQ9q5xUzwgFmVJGXiga2BeN+zpX7Vf9vmsiIa2J+MUw=="],
+
+    "@protobufjs/path": ["@protobufjs/path@1.1.2", "", {}, "sha512-6JOcJ5Tm08dOHAbdR3GrvP+yUUfkjG5ePsHYczMFLq3ZmMkAD98cDgcT2iA1lJ9NVwFd4tH/iSSoe44YWkltEA=="],
+
+    "@protobufjs/pool": ["@protobufjs/pool@1.1.0", "", {}, "sha512-0kELaGSIDBKvcgS4zkjz1PeddatrjYcmMWOlAuAPwAeccUrPHdUqo/J6LiymHHEiJT5NrF1UVwxY14f+fy4WQw=="],
+
+    "@protobufjs/utf8": ["@protobufjs/utf8@1.1.1", "", {}, "sha512-oOAWABowe8EAbMyWKM0tYDKi8Yaox52D+HWZhAIJqQXbqe0xI/GV7FhLWqlEKreMkfDjshR5FKgi3mnle0h6Eg=="],
+
+    "@silvia-odwyer/photon-node": ["@silvia-odwyer/photon-node@0.3.4", "", {}, "sha512-bnly4BKB3KDTFxrUIcgCLbaeVVS8lrAkri1pEzskpmxu9MdfGQTy8b8EgcD83ywD3RPMsIulY8xJH5Awa+t9fA=="],
+
+    "@smithy/core": ["@smithy/core@3.24.6", "", { "dependencies": { "@aws-crypto/crc32": "5.2.0", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-wBXDRup6UU97VKyaiRo8AssnfStPtG0oAAfpq/bC0a1YYau8pM86YB4kM6ccoVi1mS8l/UHbn9oDM+7uozr/ug=="],
+
+    "@smithy/credential-provider-imds": ["@smithy/credential-provider-imds@4.3.8", "", { "dependencies": { "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-5cAM+KZC02sTqDt6NaLXyu50M/GNMd1eTzDVR8Lb0BBsVtu7RWHo47VPPEEv1vt3Yub6uzr+M5FHC+GtoT0USg=="],
+
+    "@smithy/fetch-http-handler": ["@smithy/fetch-http-handler@5.4.6", "", { "dependencies": { "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-FEwEYJ1jlBKdhe9TPzfghEi1bP55ZeEImlDkEa62bBBYzUcnB6RUCyuiS2mqKt6ZVjUbBgcNhzfIctH+Hevx9g=="],
+
+    "@smithy/is-array-buffer": ["@smithy/is-array-buffer@2.2.0", "", { "dependencies": { "tslib": "^2.6.2" } }, "sha512-GGP3O9QFD24uGeAXYUjwSTXARoqpZykHadOmA8G5vfJPK0/DC67qa//0qvqrJzL1xc8WQWX7/yc7fwudjPHPhA=="],
+
+    "@smithy/node-http-handler": ["@smithy/node-http-handler@4.7.3", "", { "dependencies": { "@smithy/core": "^3.24.3", "@smithy/types": "^4.14.2", "tslib": "^2.6.2" } }, "sha512-/jPhevcTFPMVl6KNjbaI47iOg1zxC7IsnX4PQDGVZKMFceOXtB8IEYaB7a9VvkP/3oC60WzTeKocvSI7vLT0vA=="],
+
+    "@smithy/signature-v4": ["@smithy/signature-v4@5.4.6", "", { "dependencies": { "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-Ojg4B6oIDlIr1R86xCDJt1zJWnYa0VINmqdjfe9qxWjdRivHalZ3iSlQgVqYbW0MdpFOC5XfHEWsnbmdnpIILQ=="],
+
+    "@smithy/types": ["@smithy/types@4.14.3", "", { "dependencies": { "tslib": "^2.6.2" } }, "sha512-YupL0ZWmFtJexUN2cHzkvvF/b9pKrtAIfT1o7/oY/Ppu8IYeZ+lDPM5vZdQJaSeA132dJCqojjGC9NhXeF71VQ=="],
+
+    "@smithy/util-buffer-from": ["@smithy/util-buffer-from@2.2.0", "", { "dependencies": { "@smithy/is-array-buffer": "^2.2.0", "tslib": "^2.6.2" } }, "sha512-IJdWBbTcMQ6DA0gdNhh/BwrLkDR+ADW5Kr1aZmd4k3DIF6ezMV4R2NIAmT08wQJ3yUK82thHWmC/TnK/wpMMIA=="],
+
+    "@smithy/util-utf8": ["@smithy/util-utf8@2.3.0", "", { "dependencies": { "@smithy/util-buffer-from": "^2.2.0", "tslib": "^2.6.2" } }, "sha512-R8Rdn8Hy72KKcebgLiv8jQcQkXoLMOGGv5uI1/k0l+snqkOzQ1R0ChUBCxWMlBsFMekWjq0wRudIweFs7sKT5A=="],
+
+    "@types/node": ["@types/node@20.19.42", "", { "dependencies": { "undici-types": "~6.21.0" } }, "sha512-5L7SUaFC1RyDraj2yRhyBzHTobyXHmohD100CChNtyPyleoq37Mqab5Gn8XEKI04dfN/oqPdpHk38MgcQWHbZg=="],
+
+    "@types/retry": ["@types/retry@0.12.0", "", {}, "sha512-wWKOClTTiizcZhXnPY4wikVAwmdYHp8q6DmC+EJUzAMsycb7HB32Kh9RN4+0gExjmPmZSAQjgURXIGATPegAvA=="],
+
+    "agent-base": ["agent-base@7.1.4", "", {}, "sha512-MnA+YT8fwfJPgBx3m60MNqakm30XOkyIoH1y6huTQvC0PwZG7ki8NacLBcrPbNoo8vEZy7Jpuk7+jMO+CUovTQ=="],
+
+    "balanced-match": ["balanced-match@4.0.4", "", {}, "sha512-BLrgEcRTwX2o6gGxGOCNyMvGSp35YofuYzw9h1IMTRmKqttAZZVU67bdb9Pr2vUHA8+j3i2tJfjO6C6+4myGTA=="],
+
+    "base64-js": ["base64-js@1.5.1", "", {}, "sha512-AKpaYlHn8t4SVbOHCy+b5+KKgvR4vrsD8vbvrbiQJps7fKDTkjkDry6ji0rUJjC0kzbNePLwzxq8iypo41qeWA=="],
+
+    "bignumber.js": ["bignumber.js@9.3.1", "", {}, "sha512-Ko0uX15oIUS7wJ3Rb30Fs6SkVbLmPBAKdlm7q9+ak9bbIeFf0MwuBsQV6z7+X768/cHsfg+WlysDWJcmthjsjQ=="],
+
+    "bowser": ["bowser@2.14.1", "", {}, "sha512-tzPjzCxygAKWFOJP011oxFHs57HzIhOEracIgAePE4pqB3LikALKnSzUyU4MGs9/iCEUuHlAJTjTc5M+u7YEGg=="],
+
+    "brace-expansion": ["brace-expansion@5.0.6", "", { "dependencies": { "balanced-match": "^4.0.2" } }, "sha512-kLpxurY4Z4r9sgMsyG0Z9uzsBlgiU/EFKhj/h91/8yHu0edo7XuixOIH3VcJ8kkxs6/jPzoI6U9Vj3WqbMQ94g=="],
+
+    "buffer-equal-constant-time": ["buffer-equal-constant-time@1.0.1", "", {}, "sha512-zRpUiDwd/xk6ADqPMATG8vc9VPrkck7T07OIx0gnjmJAnHnTVXNQG3vfvWNuiZIkwu9KrKdA1iJKfsfTVxE6NA=="],
+
+    "bun-types": ["bun-types@1.3.14", "", { "dependencies": { "@types/node": "*" } }, "sha512-4N0ig0fEomHt5R0KCFWjovxow98rIoRwKolrYdCcknNwMekCXRnWEUvgu5soYV8QXtVsrUD8B95MBOZGPvr6KQ=="],
+
+    "chalk": ["chalk@5.6.2", "", {}, "sha512-7NzBL0rN6fMUW+f7A6Io4h40qQlG+xGmtMxfbnH/K7TAtt8JQWVQK+6g0UXKMeVJoyV5EkkNsErQ8pVD3bLHbA=="],
+
+    "cross-spawn": ["cross-spawn@7.0.6", "", { "dependencies": { "path-key": "^3.1.0", "shebang-command": "^2.0.0", "which": "^2.0.1" } }, "sha512-uV2QOWP2nWzsy2aMp8aRibhi9dlzF5Hgh5SHaB9OiTGEyDTiJJyx0uy51QXdyWbtAHNua4XJzUKca3OzKUd3vA=="],
+
+    "data-uri-to-buffer": ["data-uri-to-buffer@4.0.1", "", {}, "sha512-0R9ikRb668HB7QDxT1vkpuUBtqc53YyAwMwGeUFKRojY/NWKvdZ+9UYtRfGmhqNbRkTSVpMbmyhXipFFv2cb/A=="],
+
+    "debug": ["debug@4.4.3", "", { "dependencies": { "ms": "^2.1.3" } }, "sha512-RGwwWnwQvkVfavKVt22FGLw+xYSdzARwm0ru6DhTVA3umU5hZc28V3kO4stgYryrTlLpuvgI9GiijltAjNbcqA=="],
+
+    "diff": ["diff@8.0.4", "", {}, "sha512-DPi0FmjiSU5EvQV0++GFDOJ9ASQUVFh5kD+OzOnYdi7n3Wpm9hWWGfB/O2blfHcMVTL5WkQXSnRiK9makhrcnw=="],
+
+    "ecdsa-sig-formatter": ["ecdsa-sig-formatter@1.0.11", "", { "dependencies": { "safe-buffer": "^5.0.1" } }, "sha512-nagl3RYrbNv6kQkeJIpt6NJZy8twLB/2vtz6yN9Z4vRKHN4/QZJIEbqohALSgwKdnksuY3k5Addp5lg8sVoVcQ=="],
+
+    "extend": ["extend@3.0.2", "", {}, "sha512-fjquC59cD7CyW6urNXK0FBufkZcoiGG80wTuPujX590cB5Ttln20E2UB4S/WARVqhXffZl2LNgS+gQdPIIim/g=="],
+
+    "fast-xml-builder": ["fast-xml-builder@1.2.0", "", { "dependencies": { "path-expression-matcher": "^1.5.0", "xml-naming": "^0.1.0" } }, "sha512-00aAWieqff+ZJhsXA4g1g7M8k+7AYoMUUHF+/zFb5U6Uv/P0Vl4QZo84/IcufzYalLuEj9928bXN9PbbFzMF0Q=="],
+
+    "fast-xml-parser": ["fast-xml-parser@5.7.3", "", { "dependencies": { "@nodable/entities": "^2.1.0", "fast-xml-builder": "^1.1.7", "path-expression-matcher": "^1.5.0", "strnum": "^2.2.3" }, "bin": { "fxparser": "src/cli/cli.js" } }, "sha512-C0AaNuC+mscy6vrAQKAc/rMq+zAPHodfHGZu4sGVehvAQt/JLG1O5zEcYcXSY5zSqr4YVgxsB+pHXTq0i7eDlg=="],
+
+    "fetch-blob": ["fetch-blob@3.2.0", "", { "dependencies": { "node-domexception": "^1.0.0", "web-streams-polyfill": "^3.0.3" } }, "sha512-7yAQpD2UMJzLi1Dqv7qFYnPbaPx7ZfFK6PiIxQ4PfkGPyNyl2Ugx+a/umUonmKqjhM4DnfbMvdX6otXq83soQQ=="],
+
+    "formdata-polyfill": ["formdata-polyfill@4.0.10", "", { "dependencies": { "fetch-blob": "^3.1.2" } }, "sha512-buewHzMvYL29jdeQTVILecSaZKnt/RJWjoZCF5OW60Z67/GmSLBkOFM7qh1PI3zFNtJbaZL5eQu1vLfazOwj4g=="],
+
+    "gaxios": ["gaxios@7.1.5", "", { "dependencies": { "extend": "^3.0.2", "https-proxy-agent": "^7.0.1", "node-fetch": "^3.3.2" } }, "sha512-5FZy72Rh8LhtjmvDrKkI+lVhrsQrVKVsItxMoDm5mNQE+xR0WVIIs+jzPSJgBvKVsLi24fZhXJIsNI0bihDzFg=="],
+
+    "gcp-metadata": ["gcp-metadata@8.1.2", "", { "dependencies": { "gaxios": "^7.0.0", "google-logging-utils": "^1.0.0", "json-bigint": "^1.0.0" } }, "sha512-zV/5HKTfCeKWnxG0Dmrw51hEWFGfcF2xiXqcA3+J90WDuP0SvoiSO5ORvcBsifmx/FoIjgQN3oNOGaQ5PhLFkg=="],
+
+    "get-east-asian-width": ["get-east-asian-width@1.6.0", "", {}, "sha512-QRbvDIbx6YklUe6RxeTeleMR0yv3cYH6PsPZHcnVn7xv7zO1BHN8r0XETu8n6Ye3Q+ahtSarc3WgtNWmehIBfA=="],
+
+    "glob": ["glob@13.0.6", "", { "dependencies": { "minimatch": "^10.2.2", "minipass": "^7.1.3", "path-scurry": "^2.0.2" } }, "sha512-Wjlyrolmm8uDpm/ogGyXZXb1Z+Ca2B8NbJwqBVg0axK9GbBeoS7yGV6vjXnYdGm6X53iehEuxxbyiKp8QmN4Vw=="],
+
+    "google-auth-library": ["google-auth-library@10.7.0", "", { "dependencies": { "base64-js": "^1.3.0", "ecdsa-sig-formatter": "^1.0.11", "gaxios": "^7.1.4", "gcp-metadata": "8.1.2", "google-logging-utils": "1.1.3", "jws": "^4.0.0" } }, "sha512-QpTAbNJ36TliZLx3TTtahR8HG0hN9RllL1e3FymOvQSIKK8JmgV58H924ub2wa2DsS3ANjjP1Aw1N+Ramc8hqQ=="],
+
+    "google-logging-utils": ["google-logging-utils@1.1.3", "", {}, "sha512-eAmLkjDjAFCVXg7A1unxHsLf961m6y17QFqXqAXGj/gVkKFrEICfStRfwUlGNfeCEjNRa32JEWOUTlYXPyyKvA=="],
+
+    "graceful-fs": ["graceful-fs@4.2.11", "", {}, "sha512-RbJ5/jmFcNNCcDV5o9eTnBLJ/HszWV0P73bc+Ff4nS/rJj+YaS6IGyiOL0VoBYX+l1Wrl3k63h/KrH+nhJ0XvQ=="],
+
+    "highlight.js": ["highlight.js@10.7.3", "", {}, "sha512-tzcUFauisWKNHaRkN4Wjl/ZA07gENAjFl3J/c480dprkGTg5EQstgaNFqBfUqCq54kZRIEcreTsAgF/m2quD7A=="],
+
+    "hosted-git-info": ["hosted-git-info@9.0.3", "", { "dependencies": { "lru-cache": "^11.1.0" } }, "sha512-Hc+ghLoSt6QaYZUv0WBiIvmMDZuZZ7oaDvdH8MbfOO4lOsxdXLEvuC6ePoGs9H1X9oCLyq6+NVN0MKqD+ydxyg=="],
+
+    "http-proxy-agent": ["http-proxy-agent@7.0.2", "", { "dependencies": { "agent-base": "^7.1.0", "debug": "^4.3.4" } }, "sha512-T1gkAiYYDWYx3V5Bmyu7HcfcvL7mUrTWiM6yOfa3PIphViJ/gFPbvidQ+veqSOHci/PxBcDabeUNCzpOODJZig=="],
+
+    "https-proxy-agent": ["https-proxy-agent@7.0.6", "", { "dependencies": { "agent-base": "^7.1.2", "debug": "4" } }, "sha512-vK9P5/iUfdl95AI+JVyUuIcVtd4ofvtrOr3HNtM2yxC9bnMbEdp3x01OhQNnjb8IJYi38VlTE3mBXwcfvywuSw=="],
+
+    "ignore": ["ignore@7.0.5", "", {}, "sha512-Hs59xBNfUIunMFgWAbGX5cq6893IbWg4KnrjbYwX3tx0ztorVgTDA6B2sxf8ejHJ4wz8BqGUMYlnzNBer5NvGg=="],
+
+    "isexe": ["isexe@2.0.0", "", {}, "sha512-RHxMLp9lnKHGHRng9QFhRCMbYAcVpn69smSGcq3f36xjgVVWThj4qqLbTLlq7Ssj8B+fIQ1EuCEGI2lKsyQeIw=="],
+
+    "jiti": ["jiti@2.7.0", "", { "bin": { "jiti": "lib/jiti-cli.mjs" } }, "sha512-AC/7JofJvZGrrneWNaEnJeOLUx+JlGt7tNa0wZiRPT4MY1wmfKjt2+6O2p2uz2+skll8OZZmJMNqeke7kKbNgQ=="],
+
+    "json-bigint": ["json-bigint@1.0.0", "", { "dependencies": { "bignumber.js": "^9.0.0" } }, "sha512-SiPv/8VpZuWbvLSMtTDU8hEfrZWg/mH/nV/b4o0CYbSxu1UIQPLdwKOCIyLQX+VIPO5vrLX3i8qtqFyhdPSUSQ=="],
+
+    "json-schema-to-ts": ["json-schema-to-ts@3.1.1", "", { "dependencies": { "@babel/runtime": "^7.18.3", "ts-algebra": "^2.0.0" } }, "sha512-+DWg8jCJG2TEnpy7kOm/7/AxaYoaRbjVB4LFZLySZlWn8exGs3A4OLJR966cVvU26N7X9TWxl+Jsw7dzAqKT6g=="],
+
+    "jwa": ["jwa@2.0.1", "", { "dependencies": { "buffer-equal-constant-time": "^1.0.1", "ecdsa-sig-formatter": "1.0.11", "safe-buffer": "^5.0.1" } }, "sha512-hRF04fqJIP8Abbkq5NKGN0Bbr3JxlQ+qhZufXVr0DvujKy93ZCbXZMHDL4EOtodSbCWxOqR8MS1tXA5hwqCXDg=="],
+
+    "jws": ["jws@4.0.1", "", { "dependencies": { "jwa": "^2.0.1", "safe-buffer": "^5.0.1" } }, "sha512-EKI/M/yqPncGUUh44xz0PxSidXFr/+r0pA70+gIYhjv+et7yxM+s29Y+VGDkovRofQem0fs7Uvf4+YmAdyRduA=="],
+
+    "long": ["long@5.3.2", "", {}, "sha512-mNAgZ1GmyNhD7AuqnTG3/VQ26o760+ZYBPKjPvugO8+nLbYfX6TVpJPseBvopbdY+qpZ/lKUnmEc1LeZYS3QAA=="],
+
+    "lru-cache": ["lru-cache@11.5.1", "", {}, "sha512-RPimw/7aMdv2oqRrxKwvZXcPfwBrn/JZ2xYcY9Hus/6LaS3VOAKVWKWgNLCFSiOm1ESXinjsDlidVU7JlnCN2A=="],
+
+    "marked": ["marked@15.0.12", "", { "bin": { "marked": "bin/marked.js" } }, "sha512-8dD6FusOQSrpv9Z1rdNMdlSgQOIP880DHqnohobOmYLElGEqAL/JvxvuxZO16r4HtjTlfPRDC1hbvxC9dPN2nA=="],
+
+    "minimatch": ["minimatch@10.2.5", "", { "dependencies": { "brace-expansion": "^5.0.5" } }, "sha512-MULkVLfKGYDFYejP07QOurDLLQpcjk7Fw+7jXS2R2czRQzR56yHRveU5NDJEOviH+hETZKSkIk5c+T23GjFUMg=="],
+
+    "minipass": ["minipass@7.1.3", "", {}, "sha512-tEBHqDnIoM/1rXME1zgka9g6Q2lcoCkxHLuc7ODJ5BxbP5d4c2Z5cGgtXAku59200Cx7diuHTOYfSBD8n6mm8A=="],
+
+    "ms": ["ms@2.1.3", "", {}, "sha512-6FlzubTLZG3J2a/NVCAleEhjzq5oxgHyaCU9yYXvcLsvoVaHJq/s5xXI6/XXP6tz7R9xAOtHnSO/tXtF3WRTlA=="],
+
+    "node-domexception": ["node-domexception@1.0.0", "", {}, "sha512-/jKZoMpw0F8GRwl4/eLROPA3cfcXtLApP0QzLmUT/HuPCZWyB7IY9ZrMeKw2O/nFIqPQB3PVM9aYm0F312AXDQ=="],
+
+    "node-fetch": ["node-fetch@3.3.2", "", { "dependencies": { "data-uri-to-buffer": "^4.0.0", "fetch-blob": "^3.1.4", "formdata-polyfill": "^4.0.10" } }, "sha512-dRB78srN/l6gqWulah9SrxeYnxeddIG30+GOqK/9OlLVyLg3HPnr6SqOWTWOXKRwC2eGYCkZ59NNuSgvSrpgOA=="],
+
+    "openai": ["openai@6.26.0", "", { "peerDependencies": { "ws": "^8.18.0", "zod": "^3.25 || ^4.0" }, "optionalPeers": ["ws", "zod"], "bin": { "openai": "bin/cli" } }, "sha512-zd23dbWTjiJ6sSAX6s0HrCZi41JwTA1bQVs0wLQPZ2/5o2gxOJA5wh7yOAUgwYybfhDXyhwlpeQf7Mlgx8EOCA=="],
+
+    "p-retry": ["p-retry@4.6.2", "", { "dependencies": { "@types/retry": "0.12.0", "retry": "^0.13.1" } }, "sha512-312Id396EbJdvRONlngUx0NydfrIQ5lsYu0znKVUzVvArzEIt08V1qhtyESbGVd1FGX7UKtiFp5uwKZdM8wIuQ=="],
+
+    "partial-json": ["partial-json@0.1.7", "", {}, "sha512-Njv/59hHaokb/hRUjce3Hdv12wd60MtM9Z5Olmn+nehe0QDAsRtRbJPvJ0Z91TusF0SuZRIvnM+S4l6EIP8leA=="],
+
+    "path-expression-matcher": ["path-expression-matcher@1.5.0", "", {}, "sha512-cbrerZV+6rvdQrrD+iGMcZFEiiSrbv9Tfdkvnusy6y0x0GKBXREFg/Y65GhIfm0tnLntThhzCnfKwp1WRjeCyQ=="],
+
+    "path-key": ["path-key@3.1.1", "", {}, "sha512-ojmeN0qd+y0jszEtoY48r0Peq5dwMEkIlCOu6Q5f41lfkswXuKtYrhgoTpLnyIcHm24Uhqx+5Tqm2InSwLhE6Q=="],
+
+    "path-scurry": ["path-scurry@2.0.2", "", { "dependencies": { "lru-cache": "^11.0.0", "minipass": "^7.1.2" } }, "sha512-3O/iVVsJAPsOnpwWIeD+d6z/7PmqApyQePUtCndjatj/9I5LylHvt5qluFaBT3I5h3r1ejfR056c+FCv+NnNXg=="],
+
+    "proper-lockfile": ["proper-lockfile@4.1.2", "", { "dependencies": { "graceful-fs": "^4.2.4", "retry": "^0.12.0", "signal-exit": "^3.0.2" } }, "sha512-TjNPblN4BwAWMXU8s9AEz4JmQxnD1NNL7bNOY/AKUzyamc379FWASUhc/K1pL2noVb+XmZKLL68cjzLsiOAMaA=="],
+
+    "protobufjs": ["protobufjs@7.6.2", "", { "dependencies": { "@protobufjs/aspromise": "^1.1.2", "@protobufjs/base64": "^1.1.2", "@protobufjs/codegen": "^2.0.5", "@protobufjs/eventemitter": "^1.1.1", "@protobufjs/fetch": "^1.1.1", "@protobufjs/float": "^1.0.2", "@protobufjs/inquire": "^1.1.2", "@protobufjs/path": "^1.1.2", "@protobufjs/pool": "^1.1.0", "@protobufjs/utf8": "^1.1.1", "@types/node": ">=13.7.0", "long": "^5.3.2" } }, "sha512-N9EiLovGEQOJSPF26Ij7qUGvahfEnq0eeYZ02aigIedkmz1qZSwjnP9SBITHJuF/6MYbIW4HDN8zdYjsjqJKXQ=="],
+
+    "retry": ["retry@0.12.0", "", {}, "sha512-9LkiTwjUh6rT555DtE9rTX+BKByPfrMzEAtnlEtdEwr3Nkffwiihqe2bWADg+OQRjt9gl6ICdmB/ZFDCGAtSow=="],
+
+    "safe-buffer": ["safe-buffer@5.2.1", "", {}, "sha512-rp3So07KcdmmKbGvgaNxQSJr7bGVSVk5S9Eq1F+ppbRo70+YeaDxkw5Dd8NPN+GD6bjnYm2VuPuCXmpuYvmCXQ=="],
+
+    "shebang-command": ["shebang-command@2.0.0", "", { "dependencies": { "shebang-regex": "^3.0.0" } }, "sha512-kHxr2zZpYtdmrN1qDjrrX/Z1rR1kG8Dx+gkpK1G4eXmvXswmcE1hTWBWYUzlraYw1/yZp6YuDY77YtvbN0dmDA=="],
+
+    "shebang-regex": ["shebang-regex@3.0.0", "", {}, "sha512-7++dFhtcx3353uBaq8DDR4NuxBetBzC7ZQOhmTQInHEd6bSrXdiEyzCvG07Z44UYdLShWUyXt5M/yhz8ekcb1A=="],
+
+    "signal-exit": ["signal-exit@3.0.7", "", {}, "sha512-wnD2ZE+l+SPC/uoS0vXeE9L1+0wuaMqKlfz9AMUo38JsyLSBWSFcHR1Rri62LZc12vLr1gb3jl7iwQhgwpAbGQ=="],
+
+    "strnum": ["strnum@2.3.0", "", {}, "sha512-ums3KNd42PGyx5xaoVTO1mjU1bH3NpY4vsrVlnv9PNGqQj8wd7rJ6nEypLrJ7z5vxK5RP0yMLo6J/Gsm62DI5Q=="],
+
+    "ts-algebra": ["ts-algebra@2.0.0", "", {}, "sha512-FPAhNPFMrkwz76P7cdjdmiShwMynZYN6SgOujD1urY4oNm80Ou9oMdmbR45LotcKOXoy7wSmHkRFE6Mxbrhefw=="],
+
+    "tslib": ["tslib@2.8.1", "", {}, "sha512-oJFu94HQb+KVduSUQL7wnpmqnfmLsOA/nAh6b6EH0wCEoK0/mPeXU6c3wKDV83MkOuHPRHtSXKKU99IBazS/2w=="],
+
+    "typebox": ["typebox@1.1.38", "", {}, "sha512-pZ0aQPmMmXoUvSbeuWf/Hzsc+avNw/Zd6VeE8CFgkVGWyuHPJvqeJJDeJqLve+K70LvjYIoleGcoJHPT17cWoA=="],
+
+    "typescript": ["typescript@5.9.3", "", { "bin": { "tsc": "bin/tsc", "tsserver": "bin/tsserver" } }, "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw=="],
+
+    "undici": ["undici@8.3.0", "", {}, "sha512-TkUDgb6tl7KOGZ+7e8E3d2FYgUQgF6z5YypqjWmixVQSQERFcVrVg0ySADm2LVLRh5ljAaHTCR5Fmz3Q34rB7Q=="],
+
+    "undici-types": ["undici-types@6.21.0", "", {}, "sha512-iwDZqg0QAGrg9Rav5H4n0M64c3mkR59cJ6wQp+7C4nI0gsmExaedaYLNO44eT4AtBBwjbTiGPMlt2Md0T9H9JQ=="],
+
+    "web-streams-polyfill": ["web-streams-polyfill@3.3.3", "", {}, "sha512-d2JWLCivmZYTSIoge9MsgFCZrt571BikcWGYkjC1khllbTeDlGqZ2D8vD8E/lJa8WGWbb7Plm8/XJYV7IJHZZw=="],
+
+    "which": ["which@2.0.2", "", { "dependencies": { "isexe": "^2.0.0" }, "bin": { "node-which": "./bin/node-which" } }, "sha512-BLI3Tl1TW3Pvl70l3yq3Y64i+awpwXqsGBYWkkqMtnbXgrMD+yj7rhW0kuEDxzJaYXGjEW5ogapKNMEKNMjibA=="],
+
+    "ws": ["ws@8.21.0", "", { "peerDependencies": { "bufferutil": "^4.0.1", "utf-8-validate": ">=5.0.2" }, "optionalPeers": ["bufferutil", "utf-8-validate"] }, "sha512-Vsp28b7DRcimFQvrqu2Wek3z1iYxDCWqHYB8Qsnk/S4RfaCQzPGPyBNuVjJV3cd6UiKtUtp6sNM77gWvzcCH+g=="],
+
+    "xml-naming": ["xml-naming@0.1.0", "", {}, "sha512-k8KO9hrMyNk6tUWqUfkTEZbezRRpONVOzUTnc97VnCvyj6Tf9lyUR9EDAIeiVLv56jsMcoXEwjW8Kv5yPY52lw=="],
+
+    "yaml": ["yaml@2.9.0", "", { "bin": { "yaml": "bin.mjs" } }, "sha512-2AvhNX3mb8zd6Zy7INTtSpl1F15HW6Wnqj0srWlkKLcpYl/gMIMJiyuGq2KeI2YFxUPjdlB+3Lc10seMLtL4cA=="],
+
+    "zod": ["zod@4.4.3", "", {}, "sha512-ytENFjIJFl2UwYglde2jchW2Hwm4GJFLDiSXWdTrJQBIN9Fcyp7n4DhxJEiWNAJMV1/BqWfW/kkg71UDcHJyTQ=="],
+
+    "zod-to-json-schema": ["zod-to-json-schema@3.25.2", "", { "peerDependencies": { "zod": "^3.25.28 || ^4" } }, "sha512-O/PgfnpT1xKSDeQYSCfRI5Gy3hPf91mKVDuYLUHZJMiDFptvP41MSnWofm8dnCm0256ZNfZIM7DSzuSMAFnjHA=="],
+
+    "@aws-sdk/credential-provider-http/@smithy/node-http-handler": ["@smithy/node-http-handler@4.7.7", "", { "dependencies": { "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-ZAFvHXrEk6K180EVhmZVg8GU5pUH5BSFqRs27JW3j1qEFx9YyYwWFx17x/MHcjALYimGAji7qEOlF1++be+G5A=="],
+
+    "@aws-sdk/credential-provider-sso/@aws-sdk/token-providers": ["@aws-sdk/token-providers@3.1064.0", "", { "dependencies": { "@aws-sdk/core": "^3.974.19", "@aws-sdk/nested-clients": "^3.997.18", "@aws-sdk/types": "^3.973.12", "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-sjI+iA4JtgeckBgKwPQF7KzWillRoNDmtpiM0TRa0syiAKFHKUSf84kPXSO3+gA7aMMSxrcxzOM2oPSecaJvEA=="],
+
+    "@aws-sdk/nested-clients/@smithy/node-http-handler": ["@smithy/node-http-handler@4.7.7", "", { "dependencies": { "@smithy/core": "^3.24.6", "@smithy/types": "^4.14.3", "tslib": "^2.6.2" } }, "sha512-ZAFvHXrEk6K180EVhmZVg8GU5pUH5BSFqRs27JW3j1qEFx9YyYwWFx17x/MHcjALYimGAji7qEOlF1++be+G5A=="],
+
+    "p-retry/retry": ["retry@0.13.1", "", {}, "sha512-XQBQ3I8W1Cge0Seh+6gjj03LbmRFWuoszgK9ooCpwYIrhhoO80pfq4cUkU5DkknwfOfFteRwlZ56PYOGYyFWdg=="],
+  }
+}

index.ts

@@ -9,22 +9,29 @@ import { parseTaskFile, updateTaskInFile } from "./src/parser";
 import {
 	buildExecutionPlan,
 	buildSequentialPlan,
+	formatDependencyChain,
 	formatExecutionPlan,
-  getReadyTasks,
 } from "./src/dag";
 import { ProgressTracker } from "./src/progress";
 import { buildPlanPrompt } from "./src/prompts";
 import { formatReflections } from "./src/reflection";
-import { executeBatch, type SendChatMessage } from "./src/executor";
+import {
+	executeBatch,
+	SPINNER_FRAMES,
+	type SendChatMessage,
+} from "./src/executor";
 import {
 	loadConfig,
 	resolveTaskArg,
 	formatProgressStatus,
-  formatAllPRDsStatus,
 	findProgressFile,
+	writeLoopActive,
+	deleteLoopActive,
+	readLoopActive,
+	findRalpiDir,
 } from "./src/utils";
 
-const COMMANDS = ["status", "resume", "next", "reset"] as const;
+const COMMANDS = ["plan", "resume", "reset"] as const;
 
 type ExecutionMode = "parallel" | "sequential";
 
@@ -66,9 +73,10 @@ async function selectExecutionMode(
 	ctx: ExtensionContext,
 	project: import("./src/types").Project,
 	taskFile: string,
+	config: import("./src/types").RalpiConfig,
 ): Promise<ExecutionMode> {
 	const mode = await ctx.ui.select("Execution mode for this run?", [
-    "Parallel (where dependencies allow)",
+		`Parallel (where dependencies allow)[${config.execution.maxParallel} max]`,
 		"Sequential (one at a time)",
 	]);
 	const isParallel = mode?.startsWith("Parallel") ?? false;
@@ -117,13 +125,29 @@ async function executePlanBatches(
 	plan: ReturnType<typeof buildPlanByMode>,
 	project: Parameters<typeof buildExecutionPlan>[0],
 	taskFile: string,
-  config: import("./src/types").ralpiConfig,
+	config: import("./src/types").RalpiConfig,
 	progress: ProgressTracker,
 	ctx: ExtensionContext,
 	mode: ExecutionMode,
 	sendChatMessage?: SendChatMessage,
 	projectDir?: string,
 ): Promise<void> {
+	// Write loop-active marker so widgets can be re-instantiated after a reload
+	if (projectDir) {
+		const allTaskIds = plan.batches.flatMap((b) => b.tasks.map((t) => t.id));
+		writeLoopActive(projectDir, {
+			taskFile,
+			mode,
+			startedAt: new Date().toISOString(),
+			taskIds: allTaskIds,
+			prdKey: progress.getKey(),
+		});
+	}
+
+	// Track failed task IDs across batches to block downstream tasks
+	const failedTaskIds = new Set(progress.getFailedTaskIds());
+
+	try {
 		for (const batch of plan.batches) {
 			if (progress.getState().paused) {
 				ctx.ui.notify(
@@ -156,6 +180,49 @@ async function executePlanBatches(
 				const status = progress.getTaskStatus(task.id);
 				updateTaskInFile(taskFile, task.id, status);
 			}
+
+			// Update failed task IDs after batch completes
+			const newFailed = progress.getFailedTaskIds();
+			for (const id of newFailed) {
+				failedTaskIds.add(id);
+			}
+
+			// In sequential mode, stop after any failure
+			if (mode === "sequential" && failedTaskIds.size > 0) {
+				break;
+			}
+
+			// In parallel mode, rebuild the plan to filter out newly blocked tasks
+			if (mode === "parallel") {
+				// Use buildCompletedSet to include file-based [x] completions
+				// (progress.getCompletedTaskIds() only knows about tasks completed
+				// during THIS execution session — tasks that were already [x] in the
+				// file before the run started would be re-included and re-executed).
+				const completed = buildCompletedSet(progress, project);
+				const newPlan = buildExecutionPlan(
+					project,
+					completed,
+					undefined,
+					failedTaskIds,
+				);
+
+				// Keep processed batches (up to current batch), replace the rest
+				// with the fresh plan — its batchIndex restarts at 0, so filtering
+				// by batchIndex > currentIdx would incorrectly drop the next batch.
+				const processedCount = plan.batches.indexOf(batch) + 1;
+				plan.batches.length = processedCount;
+				plan.batches.push(...newPlan.batches);
+
+				// Skip if nothing remaining
+				if (plan.batches.length === processedCount) {
+					break;
+				}
+			}
+		}
+	} finally {
+		if (projectDir) {
+			deleteLoopActive(projectDir);
+		}
 	}
 }
 
@@ -220,6 +287,341 @@ export default function ralpiLoopExtension(pi: ExtensionAPI): void {
 		},
 	);
 
+	// ─── Reload detection: re-instantiate widgets when session reloads ──────
+	//
+	// When the user types /reload while ralpi tasks are executing, the old
+	// ExtensionContext is torn down and widgets (created via ctx.ui.setWidget)
+	// disappear. This handler detects the reload, reads the persisted loop-active
+	// marker and progress.json, and re-creates live-status widgets that show
+	// task progress with spinner animation and tool calls from session files.
+	pi.on("session_start", async (event, ctx) => {
+		if (event.reason !== "reload") return;
+
+		// Find the ralpi project directory
+		const projectDir = findRalpiDir(ctx.cwd);
+		if (!projectDir) return;
+
+		// Check if a task execution loop was active before the reload
+		const loopState = readLoopActive(projectDir);
+		if (!loopState) return;
+
+		// Load progress state
+		let abortPolling = false;
+		const progressPath = path.join(projectDir, ".ralpi", "progress.json");
+		const sessionsDir = path.join(projectDir, ".ralpi", "sessions");
+
+		// Parse the task file to get task titles
+		const titleMap = new Map<string, string>();
+		try {
+			const project = parseTaskFile(loopState.taskFile);
+			for (const task of project.tasks) {
+				titleMap.set(task.id, task.title);
+			}
+		} catch {
+			// If parsing fails, just use IDs without titles
+		}
+
+		/** Read recent tool calls from a task's session file. */
+		const readRecentToolCalls = (
+			taskId: string,
+			maxLines = 30,
+		): Array<{ name: string; label: string }> => {
+			try {
+				const files = fs
+					.readdirSync(sessionsDir)
+					.filter((f) => f.startsWith(taskId + "-"))
+					.sort();
+				if (files.length === 0) return [];
+				const sessionPath = path.join(sessionsDir, files[files.length - 1]);
+				const content = fs.readFileSync(sessionPath, "utf-8");
+				const lines = content
+					.split("\n")
+					.filter((l) => l.trim())
+					.slice(-maxLines);
+				const calls: Array<{ name: string; label: string }> = [];
+				for (const line of lines) {
+					try {
+						const event = JSON.parse(line);
+						if (event.type === "tool_execution_start") {
+							calls.push({
+								name: event.toolName,
+								label: formatToolLabel(event.toolName, event.args),
+							});
+						}
+					} catch {
+						// Skip malformed lines
+					}
+				}
+				return calls;
+			} catch {
+				return [];
+			}
+		};
+
+		/**
+		 * Strip control characters and newlines from a display label so it
+		 * does not break TUI layout (tree branches, text width calculation).
+		 */
+		function sanitizeLabel(s: string): string {
+			return s
+				.replace(/\r?\n/g, " ")
+				.replace(/[\x00-\x08\x0B\x0C\x0E-\x1F\x7F]/g, "")
+				.trim();
+		}
+
+		/** Format a tool call argument into a short label. */
+		function formatToolLabel(name: string, args: unknown): string {
+			const a = args as Record<string, unknown> | undefined;
+			if (!a) return name;
+			if (name === "bash")
+				return sanitizeLabel(String(a.command ?? "").slice(0, 70));
+			if (name === "write" || name === "read" || name === "edit")
+				return sanitizeLabel(String(a.path ?? "").slice(0, 60));
+			if (name === "grep")
+				return sanitizeLabel(
+					`${a.pattern ?? "?"} — ${String(a.path ?? "").slice(0, 40)}`,
+				);
+			if (name === "find")
+				return sanitizeLabel(`${a.path ?? "."} — ${a.glob ?? "*"}`);
+			if (name === "ls")
+				return sanitizeLabel(String(a.path ?? ".").slice(0, 60));
+			return name;
+		}
+
+		/** Re-read progress from disk (old tasks still writing to it). */
+		const readTasks = (): Record<string, { status: string }> | null => {
+			try {
+				const raw = fs.readFileSync(progressPath, "utf-8");
+				const parsed = JSON.parse(raw) as Record<string, any>;
+				return parsed.prds?.[loopState.prdKey]?.tasks ?? parsed.tasks ?? null;
+			} catch {
+				return null;
+			}
+		};
+
+		// Early exit: if all tasks already finished during the reload, just clean up
+		const initialTasks = readTasks();
+		if (initialTasks) {
+			const remaining = Object.values(initialTasks).filter(
+				(t) => t.status === "in_progress",
+			).length;
+			if (remaining === 0) {
+				ctx.ui.notify("All ralpi tasks completed during reload.", "info");
+				deleteLoopActive(projectDir);
+				return;
+			}
+		}
+
+		// Show a status notification for the reconnect
+		const taskCount = loopState.taskIds.length;
+		ctx.ui.notify(
+			`Reconnected to running ralpi execution (${taskCount} tasks, ${loopState.mode} mode)`,
+			"info",
+		);
+
+		// Shared state for the widget
+		let tickCount = 0;
+		const MAX_COLLAPSED = 3;
+
+		if (loopState.mode === "parallel") {
+			// ── Parallel mode: single batch widget ──
+			const widgetKey = `ralpi-parallel-reconnect-${Date.now()}`;
+			let widgetTui: { requestRender(): void } | null = null;
+
+			const buildBatchLines = (t: typeof ctx.ui.theme): string[] => {
+				const tasks = readTasks();
+				if (!tasks) return [t.fg("dim", "(waiting for progress...)")];
+
+				const lines: string[] = [];
+				// Only show tasks that have started (in_progress, completed, failed).
+				// Pending/unstarted tasks are noise after a reload.
+				const sortedIds = [...loopState.taskIds].sort().filter((id) => {
+					const info = tasks[id];
+					return info && info.status !== "pending";
+				});
+
+				// If no tasks have started yet, show nothing — polling will pick up
+				// changes within 500ms.
+				if (sortedIds.length === 0) return [t.fg("dim", "(starting tasks...)")];
+
+				for (const id of sortedIds) {
+					const info = tasks[id]!;
+					const title = titleMap.get(id);
+					const header = title ? `${id} · ${title}` : id;
+
+					// Status icon
+					if (info.status === "completed") {
+						lines.push(`${t.fg("success", "✓")} ${header}`);
+					} else if (info.status === "failed") {
+						lines.push(`${t.fg("error", "✗")} ${header}`);
+					} else if (info.status === "in_progress") {
+						const frame = t.fg(
+							"accent",
+							SPINNER_FRAMES[tickCount % SPINNER_FRAMES.length],
+						);
+						lines.push(`${frame} ${header}`);
+
+						// Show recent tool calls for active tasks
+						const toolCalls = readRecentToolCalls(id);
+						if (toolCalls.length > 0) {
+							if (toolCalls.length <= MAX_COLLAPSED) {
+								for (let i = 0; i < toolCalls.length; i++) {
+									const tc = toolCalls[i];
+									const isLast = i === toolCalls.length - 1;
+									const branch = isLast ? "  └── " : "  ├── ";
+									lines.push(
+										`${branch}${t.fg("accent", `[${tc.name}]`)} ${tc.label}`,
+									);
+								}
+							} else {
+								const shown = toolCalls.slice(-MAX_COLLAPSED);
+								const remaining = toolCalls.length - shown.length;
+								lines.push(t.fg("dim", `  ├── …${remaining} earlier`));
+								for (let i = 0; i < shown.length; i++) {
+									const tc = shown[i];
+									const isLast = i === shown.length - 1;
+									const branch = isLast ? "  └── " : "  ├── ";
+									lines.push(
+										`${branch}${t.fg("accent", `[${tc.name}]`)} ${tc.label}`,
+									);
+								}
+							}
+						}
+					}
+				}
+				return lines;
+			};
+
+			ctx.ui.setWidget(widgetKey, (tui, t) => {
+				widgetTui = tui;
+				return {
+					render: () => buildBatchLines(t),
+					invalidate: () => widgetTui?.requestRender(),
+				};
+			});
+
+			// 100ms tick: advances spinner frame every tick, refreshes
+			// progress + tool calls every 5 ticks (500ms).
+			const tickTimer = setInterval(() => {
+				if (abortPolling) return;
+				tickCount++;
+				widgetTui?.requestRender();
+
+				if (tickCount % 5 === 0) {
+					const tasks = readTasks();
+					if (!tasks) return;
+					const activeCount = Object.values(tasks).filter(
+						(t) => t.status === "in_progress",
+					).length;
+					if (activeCount === 0) {
+						clearInterval(tickTimer);
+						ctx.ui.setWidget(widgetKey, undefined);
+						deleteLoopActive(projectDir);
+					}
+				}
+			}, 100);
+
+			// Clean up timer when extension is shut down
+			pi.on("session_shutdown", () => {
+				abortPolling = true;
+				clearInterval(tickTimer);
+			});
+		} else {
+			// ── Sequential mode: per-task widget ──
+			const currentTaskId = loopState.taskIds.find((id) => {
+				const tasks = readTasks();
+				return tasks?.[id]?.status === "in_progress";
+			});
+
+			if (currentTaskId) {
+				const widgetKey = `ralpi-task-${currentTaskId}`;
+				let widgetTui: { requestRender(): void } | null = null;
+
+				const buildLines = (t: typeof ctx.ui.theme): string[] => {
+					const tasks = readTasks();
+					const info = tasks?.[currentTaskId];
+					const title = titleMap.get(currentTaskId);
+					const header = title ? `${currentTaskId} · ${title}` : currentTaskId;
+					const lines: string[] = [];
+
+					if (!info || info.status === "pending") {
+						return [t.fg("dim", "(starting task...)")];
+					}
+
+					if (info.status === "completed") {
+						lines.push(`${t.fg("success", "✓")} ${header}`);
+					} else if (info.status === "failed") {
+						lines.push(`${t.fg("error", "✗")} ${header}`);
+					} else if (info.status === "in_progress") {
+						const frame = t.fg(
+							"accent",
+							SPINNER_FRAMES[tickCount % SPINNER_FRAMES.length],
+						);
+						lines.push(`${frame} ${header}`);
+
+						// Show recent tool calls
+						const toolCalls = readRecentToolCalls(currentTaskId);
+						if (toolCalls.length > 0) {
+							const shown = toolCalls.slice(-MAX_COLLAPSED);
+							const remaining = toolCalls.length - shown.length;
+							if (remaining > 0) {
+								lines.push(t.fg("dim", `  ├── …${remaining} earlier`));
+							}
+							for (let i = 0; i < shown.length; i++) {
+								const tc = shown[i];
+								const isLast = i === shown.length - 1;
+								const branch = isLast ? "  └── " : "  ├── ";
+								lines.push(
+									`${branch}${t.fg("accent", `[${tc.name}]`)} ${tc.label}`,
+								);
+							}
+						}
+					}
+					return lines;
+				};
+
+				ctx.ui.setWidget(widgetKey, (tui, t) => {
+					widgetTui = tui;
+					return {
+						render: () => buildLines(t),
+						invalidate: () => widgetTui?.requestRender(),
+					};
+				});
+
+				const tickTimer = setInterval(() => {
+					if (abortPolling) return;
+					tickCount++;
+					widgetTui?.requestRender();
+
+					if (tickCount % 5 === 0) {
+						const tasks = readTasks();
+						if (!tasks) return;
+						const status = tasks[currentTaskId]?.status;
+						if (status !== "in_progress") {
+							clearInterval(tickTimer);
+							// Keep widget visible a moment, then clean up
+							setTimeout(() => {
+								ctx.ui.setWidget(widgetKey, undefined);
+								deleteLoopActive(projectDir);
+							}, 3000);
+						}
+					}
+				}, 100);
+
+				pi.on("session_shutdown", () => {
+					abortPolling = true;
+					clearInterval(tickTimer);
+				});
+			} else {
+				// No task actively in progress — show a "resume" hint
+				ctx.ui.notify(
+					"No running task found. Use /ralpi resume to continue execution.",
+					"warning",
+				);
+			}
+		}
+	});
+
 	pi.registerCommand("ralpi", {
 		description:
 			"Execute tasks from a task file using DAG-based dependency resolution",
@@ -248,21 +650,37 @@ export default function ralpiLoopExtension(pi: ExtensionAPI): void {
 				return handlePlan(ctx, parts);
 			}
 			if (looksLikePath(parts[0])) {
-        return handleRun(ctx, parts, sendProgress);
+				return handleRun(
+					ctx,
+					parts,
+					sendProgress,
+					ctx.model,
+					pi.getThinkingLevel(),
+				);
 			}
 
 			const command = parts[0];
 			switch (command) {
 				case "run":
-          return handleRun(ctx, parts.slice(1), sendProgress);
+					return handleRun(
+						ctx,
+						parts.slice(1),
+						sendProgress,
+						ctx.model,
+						pi.getThinkingLevel(),
+					);
 				case "plan":
-          return handlePlan(ctx, parts.slice(1));
-        case "status":
-          return handleStatus(ctx, parts.slice(1));
+					pi.sendUserMessage("@task-manager");
+					ctx.ui.notify("Opening Task Manager...", "info");
+					return;
 				case "resume":
-          return handleResume(ctx, parts.slice(1), sendProgress);
-        case "next":
-          return handleNext(ctx, parts.slice(1), sendProgress);
+					return handleResume(
+						ctx,
+						parts.slice(1),
+						sendProgress,
+						ctx.model,
+						pi.getThinkingLevel(),
+					);
 				case "reset":
 					return handleReset(ctx, parts.slice(1));
 				default: {
@@ -316,6 +734,8 @@ async function handleRun(
 	ctx: ExtensionContext,
 	args: string[],
 	sendChatMessage?: SendChatMessage,
+	parentModel?: unknown,
+	parentThinkingLevel?: unknown,
 ): Promise<void> {
 	const taskFile = resolveTaskArg(args[0] || "README.md", process.cwd());
 
@@ -323,7 +743,13 @@ async function handleRun(
 	// auto-resume instead of starting fresh
 	const existingProgress = findProgressFile(process.cwd(), taskFile);
 	if (existingProgress) {
-    return handleResume(ctx, args.slice(0, 1), sendChatMessage);
+		return handleResume(
+			ctx,
+			args.slice(0, 1),
+			sendChatMessage,
+			parentModel,
+			parentThinkingLevel,
+		);
 	}
 
 	// No existing progress for this task — check for any progress at all
@@ -336,7 +762,13 @@ async function handleRun(
 		);
 
 		if (shouldResume?.startsWith("Yes")) {
-      return handleResume(ctx, [], sendChatMessage);
+			return handleResume(
+				ctx,
+				[],
+				sendChatMessage,
+				parentModel,
+				parentThinkingLevel,
+			);
 		}
 	}
 
@@ -346,21 +778,28 @@ async function handleRun(
 
 	const project = parseTaskFile(taskFile);
 	const config = loadConfig(projectDir);
+	config.model = parentModel ?? ctx.model;
+	config.thinkingLevel = parentThinkingLevel;
 	const progress = new ProgressTracker(projectDir, taskFile);
 
-  // Set initial status
-  ctx.ui.setStatus(
-    "ralpi",
-    `Starting ${project.tasks.length} tasks from ${path.basename(taskFile)}`,
-  );
-
 	const completed = buildCompletedSet(progress, project);
-  const mode = await selectExecutionMode(ctx, project, taskFile);
+	const mode = await selectExecutionMode(ctx, project, taskFile, config);
 	const plan = buildPlanByMode(mode, project, completed);
 
-  // Show execution plan before starting so user can see batch breakdown
+	// Show dependency chain + execution plan before starting
+	const depChain = formatDependencyChain(project);
 	const formattedPlan = formatExecutionPlan(plan);
-  ctx.ui.notify(`${formattedPlan}\n\nStarting ${mode} execution...`, "info");
+	if (mode === "parallel") {
+		ctx.ui.notify(
+			`${depChain}\n\n${formattedPlan}\n\nStarting parallel execution...`,
+			"info",
+		);
+	} else {
+		ctx.ui.notify(
+			`${formattedPlan}\n\nStarting sequential execution...`,
+			"info",
+		);
+	}
 
 	await executePlanBatches(
 		plan,
@@ -387,46 +826,7 @@ async function handleRun(
 }
 
 // ─── /ralpi status ───────────────────────────────────────────────────────────
-
-async function handleStatus(
-  ctx: ExtensionContext,
-  args: string[],
-): Promise<void> {
-  if (args[0]) {
-    const taskFile = resolveTaskArg(args[0], process.cwd());
-    const existingProgress = findProgressFile(process.cwd(), taskFile);
-    if (existingProgress) {
-      const projectDir = path.dirname(path.dirname(existingProgress.path));
-      const progress = new ProgressTracker(
-        projectDir,
-        taskFile,
-        existingProgress.prdKey,
-      );
-      ctx.ui.notify(formatProgressStatus(progress.getState()), "info");
-      return;
-    }
-    // No progress yet for this task — parse and show plan instead
-    const project = parseTaskFile(taskFile);
-    ctx.ui.notify(
-      `No progress for ${path.basename(taskFile)}. ${
-        project.tasks.length
-      } tasks found.\nUse /ralpi run ${args[0]} to start.`,
-      "info",
-    );
-    return;
-  }
-
-  const found = findProgressFile(process.cwd());
-  if (!found) {
-    ctx.ui.notify(
-      "No .ralpi/progress.json found. Start with /ralpi run [task-file]",
-      "warning",
-    );
-    return;
-  }
-
-  ctx.ui.notify(formatAllPRDsStatus(found.state), "info");
-}
+// (removed — use /ralpi plan to invoke @task-manager)
 
 // ─── /ralpi resume ───────────────────────────────────────────────────────────
 
@@ -434,6 +834,8 @@ async function handleResume(
 	ctx: ExtensionContext,
 	args: string[],
 	sendChatMessage?: SendChatMessage,
+	parentModel?: unknown,
+	parentThinkingLevel?: unknown,
 ): Promise<void> {
 	let taskFile: string;
 	let projectDir: string;
@@ -473,17 +875,30 @@ async function handleResume(
 		);
 	}
 	const config = loadConfig(projectDir);
+	config.model = parentModel ?? ctx.model;
+	config.thinkingLevel = parentThinkingLevel;
 	const progress = new ProgressTracker(projectDir, taskFile, found.prdKey);
 
 	progress.setPaused(false);
 
-  // Set resume status
-  ctx.ui.setStatus("ralpi", `Resuming from ${path.basename(taskFile)}`);
-
 	const completed = buildCompletedSet(progress, project);
-  const mode = await selectExecutionMode(ctx, project, taskFile);
+	const mode = await selectExecutionMode(ctx, project, taskFile, config);
 	const plan = buildPlanByMode(mode, project, completed);
 
+	// Print remaining batches before executing
+	const formattedPlan = formatExecutionPlan(plan);
+	if (mode === "parallel") {
+		ctx.ui.notify(
+			`${formattedPlan}\n\nResuming parallel execution...`,
+			"info",
+		);
+	} else {
+		ctx.ui.notify(
+			`${formattedPlan}\n\nResuming sequential execution...`,
+			"info",
+		);
+	}
+
 	await executePlanBatches(
 		plan,
 		project,
@@ -500,85 +915,7 @@ async function handleResume(
 }
 
 // ─── /ralpi next ─────────────────────────────────────────────────────────────
-
-async function handleNext(
-  ctx: ExtensionContext,
-  args: string[],
-  sendChatMessage?: SendChatMessage,
-): Promise<void> {
-  let taskFile: string;
-  let projectDir: string;
-  let found: ReturnType<typeof findProgressFile>;
-
-  if (args[0]) {
-    taskFile = resolveTaskArg(args[0], process.cwd());
-    found = findProgressFile(process.cwd(), taskFile);
-    if (found) {
-      projectDir = path.dirname(path.dirname(found.path));
-    } else {
-      projectDir = process.cwd();
-    }
-  } else {
-    found = findProgressFile(process.cwd());
-    if (!found) {
-      ctx.ui.notify(
-        "No .ralpi/progress.json found. Start with /ralpi run [task-file]",
-        "warning",
-      );
-      return;
-    }
-    taskFile = found.state.prds
-      ? Object.values(found.state.prds)[0].sourcePath
-      : found.state.sourcePath;
-    projectDir = path.dirname(path.dirname(found.path));
-  }
-
-  const project = parseTaskFile(taskFile);
-  if (!Array.isArray(project.tasks)) {
-    throw new Error(
-      `Parsed project from ${taskFile} has invalid tasks: expected array, got ${typeof project.tasks}`,
-    );
-  }
-  const config = loadConfig(projectDir);
-  const progress = new ProgressTracker(projectDir, taskFile, found?.prdKey);
-
-  const completed = buildCompletedSet(progress, project);
-  const ready = getReadyTasks(project, completed);
-
-  if (ready.length === 0) {
-    ctx.ui.notify(
-      "No tasks ready to execute. All tasks completed or blocked.",
-      "info",
-    );
-    return;
-  }
-
-  const nextBatch = ready.slice(
-    0,
-    config.execution.maxParallel || ready.length,
-  );
-
-  for (const task of nextBatch) {
-    await executeBatch(
-      [task],
-      project,
-      config,
-      progress,
-      ctx,
-      { parallel: false },
-      sendChatMessage,
-      projectDir,
-    );
-    updateTaskInFile(taskFile, task.id, progress.getTaskStatus(task.id));
-  }
-
-  ctx.ui.notify(
-    `Executed: ${nextBatch
-      .map((t) => t.id)
-      .join(", ")}\n\n${formatProgressStatus(progress.getState())}`,
-    "info",
-  );
-}
+// (removed — use /ralpi run to execute tasks)
 
 // ─── /ralpi reset ────────────────────────────────────────────────────────────
 

package.json

@@ -1,33 +1,51 @@
 {
-  "name": "ralpi-loop",
-  "version": "1.0.0",
-  "description": "Execute tasks from task files using DAG-based dependency resolution with persistent progress tracking",
-  "main": "dist/index.js",
+  "name": "@mikefreno/ralpi",
+  "version": "0.2.5",
+  "description": "Execute tasks from task files/PRD's using DAG-based dependency resolution with persistent progress tracking",
   "keywords": [
     "pi-package",
     "pi-extension",
     "task-runner",
     "dag",
     "task-manager",
-    "ralpi-loop",
+    "ralph-loop",
     "prd"
   ],
-  "author": "",
+  "author": "Michael Freno",
   "license": "MIT",
+  "homepage": "https://github.com/mikefreno/ralpi",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/mikefreno/ralpi.git"
+  },
+  "bugs": {
+    "url": "https://github.com/mikefreno/ralpi/issues"
+  },
   "files": [
-    "dist/",
+    "index.ts",
+    "src/",
     "skills/",
     "prompts/",
-    "index.ts"
+    "README.md",
+    "LICENSE"
   ],
   "scripts": {
-    "build": "tsc",
-    "watch": "tsc --watch",
-    "prepublishOnly": "npm run build"
+    "typecheck": "tsc --noEmit",
+    "prepublishOnly": "tsc --noEmit",
+    "test": "bun test"
+  },
+  "engines": {
+    "bun": ">=1.1.0"
   },
   "pi": {
     "extensions": [
-      "./dist/index.js"
+      "./index.ts"
+    ],
+    "skills": [
+      "./skills"
+    ],
+    "prompts": [
+      "./prompts"
     ]
   },
   "dependencies": {
@@ -37,8 +55,12 @@
     "@earendil-works/pi-coding-agent": "*",
     "@earendil-works/pi-tui": "*"
   },
+  "publishConfig": {
+    "access": "public"
+  },
   "devDependencies": {
     "@types/node": "^20.0.0",
+    "bun-types": "^1.3.14",
     "typescript": "^5.3.0"
   }
 }

prompts/task-manager.md

@@ -14,6 +14,7 @@ Purpose:
 You are a Task Manager (@task-manager), an expert at breaking down complex software features into small, verifiable subtasks. Your role is to create structured task plans that enable efficient, atomic implementation work.
 
 ## Core Responsibilities
+
 - Break complex features into atomic tasks
 - Create structured directories with task files and indexes
 - Generate clear acceptance criteria and dependency mapping
@@ -22,6 +23,7 @@ You are a Task Manager (@task-manager), an expert at breaking down complex softw
 ## Mandatory Two-Phase Workflow
 
 ### Phase 1: Planning (Approval Required)
+
 When given a complex feature request:
 
 1. **Analyze the feature** to identify:
@@ -36,21 +38,27 @@ When given a complex feature request:
    - Exit criteria for feature completion
 
 3. **Present plan using this exact format:**```
+
 ## Subtask Plan
+
 feature: {kebab-case-feature-name}
 objective: {one-line description}
 
 tasks:
+
 - seq: {2-digit}, filename: {seq}-{task-description}.md, title: {clear title}
 - seq: {2-digit}, filename: {seq}-{task-description}.md, title: {clear title}
 
 dependencies:
+
 - {seq} -> {seq} (task dependencies)
 
 exit_criteria:
+
 - {specific, measurable completion criteria}
 
 Approval needed before file creation.
+
 ```
 
 4. **Wait for explicit approval** before proceeding to Phase 2.
@@ -67,6 +75,7 @@ Once approved:
 
 **Feature Index Template** (`tasks/{feature}/README.md`):
 ```
+
 # {Feature Title}
 
 Objective: {one-liner}
@@ -74,17 +83,22 @@ Objective: {one-liner}
 Status legend: [ ] todo, [~] in-progress, [x] done
 
 Tasks
+
 - [ ] {seq} — {task-description} → `{seq}-{task-description}.md`
 
 Dependencies
+
 - {seq} depends on {seq}
 
 Exit criteria
+
 - The feature is complete when {specific criteria}
+
 ```
 
 **Task File Template** (`{seq}-{task-description}.md`):
 ```
+
 # {seq}. {Title}
 
 meta:
@@ -95,40 +109,54 @@ meta:
   tags: [implementation, tests-required]
 
 objective:
+
 - Clear, single outcome for this task
 
 deliverables:
+
 - What gets added/changed (files, modules, endpoints)
 
 steps:
+
 - Step-by-step actions to complete the task
 
 tests:
+
 - Unit: which functions/modules to cover (Arrange–Act–Assert)
 - Integration/e2e: how to validate behavior
 
 acceptance_criteria:
+
 - Observable, binary pass/fail conditions
 
 validation:
+
 - Commands or scripts to run and how to verify
 
 notes:
+
 - Assumptions, links to relevant docs or design
+
 ```
 
 3. **Provide creation summary:**
 ```
+
 ## Subtasks Created
+
 - tasks/{feature}/README.md
 - tasks/{feature}/{seq}-{task-description}.md
 
 Next suggested task: {seq} — {title}
+
 ```
 
 ## Strict Conventions
 - **Naming:** Always use kebab-case for features and task descriptions
-- **Sequencing:** 2-digits (01, 02, 03...)
+- **Sequencing:** 2-digits (01, 02, 03...) — optionally a single lowercase letter
+  suffix may be appended to insert a sub-task between two numbered steps without
+  renumbering siblings (e.g. `02b`, `02c` for sub-tasks of `02`). The parser
+  normalizes `2b` → `02b`.
 - **File pattern:** `{seq}-{task-description}.md`
 - **Dependencies:** Always map task relationships (if applicable)
 - **Tests:** Every task must include test requirements

src/dag.ts

@@ -1,4 +1,38 @@
-import type { Task, ExecutionBatch, ExecutionPlan, Project } from "./types";
+import type {
+	Task,
+	ExecutionBatch,
+	ExecutionPlan,
+	Project,
+	ParallelGroup,
+} from "./types";
+
+// ─── Blocked Tasks ───────────────────────────────────────────────────────────
+
+/**
+ * Find tasks that are blocked (direct or transitive) due to failed dependencies.
+ * Returns a Set of blocked task IDs.
+ */
+export function getBlockedTasks(
+	pendingTasks: Task[],
+	failedTaskIds: Set<string>,
+): Set<string> {
+	const blocked = new Set<string>();
+
+	let changed = true;
+	while (changed) {
+		changed = false;
+		for (const task of pendingTasks) {
+			if (blocked.has(task.id)) continue;
+			const deps = task.dependencies || [];
+			if (deps.some((dep) => failedTaskIds.has(dep) || blocked.has(dep))) {
+				blocked.add(task.id);
+				changed = true;
+			}
+		}
+	}
+
+	return blocked;
+}
 
 // ─── Main Entry ──────────────────────────────────────────────────────────────
 
@@ -10,26 +44,46 @@ export function buildExecutionPlan(
 	project: Project,
 	completed: Set<string>,
 	parallelGroup?: number,
+	failedTaskIds: Set<string> = new Set(),
 ): ExecutionPlan {
-	const allTasks = new Map(project.tasks.map((t) => [t.id, t]));
+	// Filter out already completed AND failed tasks
+	// Failed tasks should not be re-scheduled — they're only re-attempted
+	// via the retry mechanism inside executeTask, not via the DAG.
+	const pendingTasks = project.tasks.filter(
+		(t) => !completed.has(t.id) && !failedTaskIds.has(t.id),
+	);
+	const skippedTasks = project.tasks.filter(
+		(t) => completed.has(t.id) || failedTaskIds.has(t.id),
+	);
 
-	// Filter out already completed tasks
-	const pendingTasks = project.tasks.filter((t) => !completed.has(t.id));
+	// With explicitly declared parallel groups, all groups are independent.
+	// Since there are no cross-group dependencies by definition, standard
+	// Kahn's algorithm produces the correct plan — tasks ready in any group
+	// appear in the same batch, and intra-group dependencies (e.g. "21 must
+	// be done before 22, 23, 24") are respected automatically.
+	// The parallel groups are preserved as metadata for display/documentation.
+	if (project.parallelGroups && project.parallelGroups.length > 0) {
+		return {
+			batches: buildGroupAwareBatches(project, pendingTasks, failedTaskIds),
+			totalTasks: pendingTasks.length,
+			skippedTasks,
+		};
+	}
 
-	// If parallel_group is explicitly set, use group-based batching
+	// If parallel_group is explicitly set (legacy config flag), use group-based batching
 	if (parallelGroup !== undefined) {
 		return {
-			batches: buildParallelGroupBatches(pendingTasks, allTasks, completed),
+			batches: buildParallelGroupBatchesLegacy(pendingTasks, failedTaskIds),
 			totalTasks: pendingTasks.length,
-			skippedTasks: project.tasks.filter((t) => completed.has(t.id)),
+			skippedTasks,
 		};
 	}
 
 	// Use dependency-based Kahn's algorithm
 	return {
-		batches: buildBatches(pendingTasks, allTasks, completed),
+		batches: buildBatches(pendingTasks, failedTaskIds),
 		totalTasks: pendingTasks.length,
-		skippedTasks: project.tasks.filter((t) => completed.has(t.id)),
+		skippedTasks,
 	};
 }
 
@@ -41,9 +95,18 @@ export function buildExecutionPlan(
 export function buildSequentialPlan(
 	project: Project,
 	completed: Set<string>,
+	failedTaskIds: Set<string> = new Set(),
 ): ExecutionPlan {
 	const pendingTasks = project.tasks.filter((t) => !completed.has(t.id));
-	const batches: ExecutionBatch[] = pendingTasks.map((task, i) => ({
+
+	// Mark tasks with failed dependencies as skipped
+	const blocked = getBlockedTasks(pendingTasks, failedTaskIds);
+	const skippedTasks = project.tasks.filter(
+		(t) => completed.has(t.id) || blocked.has(t.id),
+	);
+	const activeTasks = pendingTasks.filter((t) => !blocked.has(t.id));
+
+	const batches: ExecutionBatch[] = activeTasks.map((task, i) => ({
 		tasks: [task],
 		batchIndex: i,
 	}));
@@ -51,7 +114,7 @@ export function buildSequentialPlan(
 	return {
 		batches,
 		totalTasks: pendingTasks.length,
-		skippedTasks: project.tasks.filter((t) => completed.has(t.id)),
+		skippedTasks,
 	};
 }
 
@@ -59,12 +122,15 @@ export function buildSequentialPlan(
 
 function buildBatches(
 	pendingTasks: Task[],
-	allTasks: Map<string, Task>,
-	completed: Set<string>,
+	failedTaskIds: Set<string>,
 ): ExecutionBatch[] {
 	const batches: ExecutionBatch[] = [];
-	const done = new Set(completed);
-	const remaining = new Set(pendingTasks.map((t) => t.id));
+	const done = new Set<string>();
+	const blocked = getBlockedTasks(pendingTasks, failedTaskIds);
+	const pendingSet = new Set(pendingTasks.map((t) => t.id));
+	const remaining = new Set(
+		pendingTasks.filter((t) => !blocked.has(t.id)).map((t) => t.id),
+	);
 
 	while (remaining.size > 0) {
 		// Find tasks whose dependencies are all satisfied
@@ -74,7 +140,7 @@ function buildBatches(
 
 			const deps = task.dependencies || [];
 			const depsSatisfied = deps.every(
-				(dep) => done.has(dep) || !allTasks.has(dep),
+				(dep) => done.has(dep) || !pendingSet.has(dep),
 			);
 
 			if (depsSatisfied) {
@@ -100,20 +166,81 @@ function buildBatches(
 	return batches;
 }
 
-// ─── Parallel Group Batching ─────────────────────────────────────────────────
+// ─── Group-Aware Batching ────────────────────────────────────────────────────
 
 /**
- * Build batches from explicit parallel_group values.
- * Groups execute in ascending order; tasks within a group run concurrently.
+ * Build batches respecting both explicit parallel groups and intra-group
+ * dependencies. Since parallel group declarations imply no cross-group
+ * dependencies, all tasks whose dependencies are satisfied — across any
+ * group — can run concurrently in the same batch. This means groups
+ * "proceed independently" as the user specified: tasks from different
+ * groups can appear in the same batch when ready.
+ *
+ * Intra-group dependencies (e.g., "21 must be done before 22, 23, 24")
+ * are handled by Kahn's algorithm: if 21 has deps satisfied but 22 doesn't,
+ * only 21 appears in the current batch.
  */
-function buildParallelGroupBatches(
+function buildGroupAwareBatches(
+	_project: Project,
 	pendingTasks: Task[],
-	allTasks: Map<string, Task>,
-	completed: Set<string>,
+	failedTaskIds: Set<string>,
 ): ExecutionBatch[] {
+	const blocked = getBlockedTasks(pendingTasks, failedTaskIds);
+	const activeTasks = pendingTasks.filter((t) => !blocked.has(t.id));
+
+	// Standard Kahn's algorithm across ALL tasks — parallel groups are
+	// metadata for display, not scheduling constraints.
+	const pendingSet = new Set(pendingTasks.map((t) => t.id));
+	const done = new Set<string>();
+	const remaining = new Set(activeTasks.map((t) => t.id));
+	const batches: ExecutionBatch[] = [];
+
+	while (remaining.size > 0) {
+		const ready: Task[] = [];
+		for (const task of activeTasks) {
+			if (!remaining.has(task.id)) continue;
+			const deps = task.dependencies || [];
+			const depsSatisfied = deps.every(
+				(dep) => done.has(dep) || !pendingSet.has(dep),
+			);
+			if (depsSatisfied) {
+				ready.push(task);
+			}
+		}
+
+		if (ready.length === 0) {
+			throw new Error(
+				`Dependency cycle detected: ${Array.from(remaining).join(", ")}`,
+			);
+		}
+
+		batches.push({ tasks: ready, batchIndex: batches.length });
+		for (const task of ready) {
+			done.add(task.id);
+			remaining.delete(task.id);
+		}
+	}
+
+	return batches;
+}
+
+// ─── Legacy Parallel Group Batching ─────────────────────────────────────────
+
+/**
+ * Legacy: build batches from explicit parallel_group values only.
+ * Groups execute in ascending order; tasks within a group run concurrently.
+ * Does NOT respect intra-group dependencies.
+ */
+function buildParallelGroupBatchesLegacy(
+	pendingTasks: Task[],
+	failedTaskIds: Set<string>,
+): ExecutionBatch[] {
+	const blocked = getBlockedTasks(pendingTasks, failedTaskIds);
+	const activeTasks = pendingTasks.filter((t) => !blocked.has(t.id));
+
 	const groups = new Map<number, Task[]>();
 
-	for (const task of pendingTasks) {
+	for (const task of activeTasks) {
 		const group = task.parallelGroup ?? 0;
 		if (!groups.has(group)) groups.set(group, []);
 		groups.get(group)!.push(task);
@@ -121,7 +248,7 @@ function buildParallelGroupBatches(
 
 	const sortedGroups = Array.from(groups.entries()).sort((a, b) => a[0] - b[0]);
 
-	return sortedGroups.map(([groupNum, tasks], i) => ({
+	return sortedGroups.map(([_groupNum, tasks], i) => ({
 		tasks,
 		batchIndex: i,
 	}));
@@ -268,18 +395,125 @@ export function getCriticalPath(project: Project): Task[] {
 	return path;
 }
 
+// ─── Format Dependency Chain ─────────────────────────────────────────────────
+
+/**
+ * Format the dependency DAG as a tree for display.
+ * Rooted at tasks with no dependencies, showing what depends on what.
+ */
+export function formatDependencyChain(project: Project): string {
+	const taskMap = new Map(project.tasks.map((t) => [t.id, t]));
+	const lines: string[] = [];
+
+	lines.push("## Dependency Chain");
+	lines.push("");
+
+	if (project.tasks.length === 0) {
+		lines.push("(no tasks)");
+		return lines.join("\n");
+	}
+
+	// Build reverse dependency map: taskId → [dependent taskIds]
+	const dependents = new Map<string, string[]>();
+	for (const task of project.tasks) {
+		dependents.set(task.id, []);
+	}
+	for (const task of project.tasks) {
+		for (const dep of task.dependencies) {
+			if (dependents.has(dep)) {
+				dependents.get(dep)!.push(task.id);
+			}
+		}
+	}
+
+	// Root tasks: those with no dependencies
+	const roots = project.tasks.filter((t) => t.dependencies.length === 0);
+	const rendered = new Set<string>();
+
+	function renderNode(taskId: string, prefix: string, isLast: boolean): void {
+		const task = taskMap.get(taskId);
+		if (!task) return;
+
+		const alreadyRendered = rendered.has(taskId);
+		rendered.add(taskId);
+
+		const connector = prefix ? (isLast ? "└── " : "├── ") : "";
+
+		if (alreadyRendered) {
+			lines.push(`${prefix}${connector}${task.id} · ${task.title}`);
+			return;
+		}
+
+		const deps =
+			task.dependencies.length > 0
+				? ` ← needs ${task.dependencies.join(", ")}`
+				: " (root)";
+
+		lines.push(
+			`${prefix}${connector}${task.id} · ${task.title}${prefix ? "" : deps}`,
+		);
+
+		const children = (dependents.get(taskId) || [])
+			.filter((c) => c !== taskId)
+			.sort();
+
+		for (let i = 0; i < children.length; i++) {
+			const childPrefix = prefix + (isLast ? "    " : "│   ");
+			renderNode(children[i], childPrefix, i === children.length - 1);
+		}
+	}
+
+	for (let i = 0; i < roots.length; i++) {
+		renderNode(roots[i].id, "", i === roots.length - 1);
+	}
+
+	// Tasks not reached from any root (have deps but no root-traversable path)
+	const unreached = project.tasks.filter((t) => !rendered.has(t.id));
+	if (unreached.length > 0) {
+		lines.push("");
+		lines.push("Orphan tasks (dependencies not in task list):");
+		for (const t of unreached) {
+			const deps =
+				t.dependencies.length > 0
+					? ` ← needs ${t.dependencies.join(", ")}`
+					: "";
+			lines.push(`  ${t.id} · ${t.title}${deps}`);
+		}
+	}
+
+	return lines.join("\n");
+}
+
 // ─── Format Execution Plan ───────────────────────────────────────────────────
 
 /**
  * Format the execution plan for display
  */
-export function formatExecutionPlan(plan: ExecutionPlan): string {
+/**
+ * Format the execution plan for display, optionally with parallel group annotations
+ */
+export function formatExecutionPlan(
+	plan: ExecutionPlan,
+	parallelGroups?: ParallelGroup[],
+): string {
 	const lines: string[] = [];
 	lines.push("## Execution Plan");
 	lines.push("");
 	lines.push(`Total tasks: ${plan.totalTasks}`);
 	lines.push(`Batches: ${plan.batches.length}`);
 
+	// Build a lookup: taskId → group label
+	const groupLabel = new Map<string, string>();
+	if (parallelGroups) {
+		for (const g of parallelGroups) {
+			for (const id of g.taskIds) {
+				if (g.label) {
+					groupLabel.set(id, g.label);
+				}
+			}
+		}
+	}
+
 	if (plan.skippedTasks.length > 0) {
 		lines.push(
 			`Already completed: ${plan.skippedTasks.map((t) => t.id).join(", ")}`,
@@ -290,7 +524,14 @@ export function formatExecutionPlan(plan: ExecutionPlan): string {
 	for (const batch of plan.batches) {
 		lines.push(`### Batch ${batch.batchIndex + 1}`);
 		for (const task of batch.tasks) {
-			lines.push(`- ${task.id}: ${task.title}`);
+			const annotation = groupLabel.has(task.id)
+				? `  _(${groupLabel.get(task.id)})_`
+				: "";
+			const deps =
+				task.dependencies.length > 0
+					? `  ← needs ${task.dependencies.join(", ")}`
+					: "";
+			lines.push(`- ${task.id}: ${task.title}${annotation}${deps}`);
 		}
 		lines.push("");
 	}

src/executor.ts

@@ -1,3 +1,4 @@
+import { truncateToWidth } from "@earendil-works/pi-tui";
 import * as path from "node:path";
 import type { Task, Project, Reflection, ToolUsage } from "./types";
 import type { RalpiConfig } from "./types";
@@ -5,14 +6,17 @@ import type { ProgressTracker } from "./progress";
 import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { buildTaskPrompt } from "./prompts";
 import { extractReflection } from "./reflection";
-import { WidgetBatcher } from "./widget-batcher";
 import {
 	runAgentSession,
 	writeFileSafe,
 	ensureDir,
 	captureGitCommits,
+	hasUncommittedChanges,
+	getGitStatusPorcelain,
+	getGitDiff,
 	formatDuration,
 } from "./utils";
+import { updateTaskInFile } from "./parser";
 
 /** Optional callback to post a progress message into the chat history. */
 export type SendChatMessage = (
@@ -26,6 +30,110 @@ export interface ToolCallEntry {
 	label: string;
 }
 
+// ─── Widget Expand/Collapse ───────────────────────────────────────────────
+
+/** Max tool calls shown in a live widget before truncating. Widgets don't
+ *  support message-style Ctrl+O expansion (that's only for chat-history
+ *  messages rendered by registerMessageRenderer). */
+const MAX_COLLAPSED = 3;
+
+export const SPINNER_FRAMES = [
+	"⠋",
+	"⠙",
+	"⠹",
+	"⠸",
+	"⠼",
+	"⠴",
+	"⠦",
+	"⠧",
+	"⠇",
+	"⠏",
+];
+
+// ─── Model Round-Robin ─────────────────────────────────────────────────────
+
+/**
+ * Round-robin model assignment with slot reuse.
+ *
+ * With models [A, B, C] and 2 concurrent tasks, only A and B are used.
+ * Model C is only touched when a third concurrent task starts.
+ * Freed slots are reused before new slots are allocated.
+ */
+class ModelRoundRobin {
+	private models: unknown[];
+	private freeSlots: number[];
+	private nextIndex = 0;
+	private assignments = new Map<string, number>();
+
+	constructor(models: unknown[]) {
+		this.models = models;
+		this.freeSlots = [];
+	}
+
+	get length(): number {
+		return this.models.length;
+	}
+
+	assign(taskId: string): unknown {
+		let index: number;
+		if (this.freeSlots.length > 0) {
+			// Reuse a freed model slot first
+			index = this.freeSlots.shift()!;
+		} else if (this.nextIndex < this.models.length) {
+			// Allocate a new slot
+			index = this.nextIndex++;
+		} else {
+			// All models in use — wrap around
+			index = this.nextIndex % this.models.length;
+			this.nextIndex++;
+		}
+		this.assignments.set(taskId, index);
+		return this.models[index];
+	}
+
+	release(taskId: string): void {
+		const index = this.assignments.get(taskId);
+		if (index !== undefined) {
+			this.freeSlots.push(index);
+			this.freeSlots.sort((a, b) => a - b);
+			this.assignments.delete(taskId);
+		}
+	}
+
+	/**
+	 * Advance a task to the next model slot without going through freed slots.
+	 * Used for model failover — when the current model is down, skip to the
+	 * next one instead of re-assigning the same freed index.
+	 */
+	advance(taskId: string): unknown {
+		const currentIndex = this.assignments.get(taskId);
+		if (currentIndex === undefined) {
+			// No current assignment — fresh assign (fallback, shouldn't happen)
+			return this.assign(taskId);
+		}
+		// If this index was freed (e.g. from an earlier release call that raced),
+		// remove it from freeSlots so it's not handed out to another task.
+		const freeIdx = this.freeSlots.indexOf(currentIndex);
+		if (freeIdx !== -1) this.freeSlots.splice(freeIdx, 1);
+		// Advance to the next index (circular)
+		const nextIndex = (currentIndex + 1) % this.models.length;
+		this.assignments.set(taskId, nextIndex);
+		return this.models[nextIndex];
+	}
+}
+
+/** Shared state for parallel-batch widget. Each running task writes its
+ *  tool calls and spinner frame; the batch widget reads them in task-ID order. */
+interface ParallelWidgetEntry {
+	taskHeader: string;
+	frameIndex: number;
+	done: boolean;
+	success: boolean;
+	toolCalls: ToolCallEntry[];
+}
+
+type ParallelWidgetState = Map<string, ParallelWidgetEntry>;
+
 // ─── Run Single Task ────────────────────────────────────────────────────────
 
 /**
@@ -40,7 +148,9 @@ export async function runTask(
 	ctx: ExtensionContext,
 	sendChatMessage?: SendChatMessage,
 	projectDir: string = project.sourceDir,
-  batcher?: WidgetBatcher,
+	parallelState?: ParallelWidgetState,
+	assignedModel?: unknown,
+	batchRender?: () => void,
 ): Promise<{
 	success: boolean;
 	reflection?: Reflection;
@@ -48,7 +158,6 @@ export async function runTask(
 	durationMs: number;
 	toolUsage?: ToolUsage;
 	outputPreview?: string;
-  sessionFile?: string;
 	commitMessages?: string[];
 	commitSummary?: string;
 }> {
@@ -62,74 +171,93 @@ export async function runTask(
 		config.prompts.projectContext,
 	);
 
-  // Write prompt to .ralpi/ with timestamp (for debugging)
-  const ralpiDir = path.join(projectDir, ".ralpi");
-  ensureDir(ralpiDir);
-  const promptFile = path.join(ralpiDir, `prompt-${startMs}.md`);
-  writeFileSafe(promptFile, prompt);
-
-  // Footer shows just the task title (no batch prefix)
-  ctx.ui.setStatus("ralpi", task.title);
-
 	const taskHeader = `${task.id} · ${task.title}`;
 
-  // Live progress widget above the editor — animated spinner + tool call tree
-  // Using setWidget instead of setWorkingMessage because the working message area
-  // is only visible during parent agent streaming, not during extension command execution.
-  // Widget key is unique per task so parallel tasks each get their own widget.
+	// When running in parallel, all tasks share a single widget so ordering
+	// is deterministic (sorted by task ID). In sequential mode each task gets
+	// its own widget.
+	const isParallel = !!parallelState;
 	const widgetKey = `ralpi-task-${task.id}`;
-  const SPINNER_FRAMES = ["⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏"];
 	let frameIndex = 0;
-  const theme = ctx.ui.theme;
-  const MAX_COLLAPSED = 3;
-
 	const toolCalls: ToolCallEntry[] = [];
+	let widgetTui: { requestRender(): void } | null = null;
 
-  const updateWidget = () => {
-    const frame = theme.fg("accent", SPINNER_FRAMES[frameIndex]);
-    const lines = [`${frame} ${taskHeader}`];
+	if (isParallel) {
+		parallelState!.set(task.id, {
+			taskHeader,
+			frameIndex: 0,
+			done: false,
+			success: false,
+			toolCalls: [],
+		});
+	} else {
+		// Build widget lines from current state. Live widgets can't expand/collapse
+		// like chat messages, so we always truncate to MAX_COLLAPSED recent calls.
+		const truncateWidth = 74; // Account for widget container padding
+		const buildLines = (t: typeof ctx.ui.theme, width?: number): string[] => {
+			const effectiveWidth = width
+				? Math.min(width, truncateWidth)
+				: truncateWidth;
+			const frame = t.fg("accent", SPINNER_FRAMES[frameIndex]);
+			const lines = [truncateToWidth(`${frame} ${taskHeader}`, effectiveWidth)];
 
 			if (toolCalls.length > 0) {
+				if (toolCalls.length <= MAX_COLLAPSED) {
+					for (let i = 0; i < toolCalls.length; i++) {
+						const entry = toolCalls[i];
+						const isLast = i === toolCalls.length - 1;
+						const branch = isLast ? "  └── " : "  ├── ";
+						const tag = t.fg("accent", `[${entry.name}]`);
+						lines.push(
+							truncateToWidth(`${branch}${tag} ${entry.label}`, effectiveWidth),
+						);
+					}
+				} else {
 					const shown = toolCalls.slice(-MAX_COLLAPSED);
 					const remaining = toolCalls.length - shown.length;
-
-      if (remaining > 0) {
-        lines.push(theme.fg("dim", `  ├── ${remaining} more`));
-      }
-
+					lines.push(
+						truncateToWidth(
+							t.fg("dim", `  ├── …${remaining} earlier`),
+							effectiveWidth,
+						),
+					);
 					for (let i = 0; i < shown.length; i++) {
 						const entry = shown[i];
 						const isLast = i === shown.length - 1;
 						const branch = isLast ? "  └── " : "  ├── ";
-        const tag = theme.fg("accent", `[${entry.name}]`);
-        lines.push(`${branch}${tag} ${entry.label}`);
+						const tag = t.fg("accent", `[${entry.name}]`);
+						lines.push(
+							truncateToWidth(`${branch}${tag} ${entry.label}`, effectiveWidth),
+						);
 					}
 				}
-
-    if (batcher) {
-      batcher.schedule(widgetKey, lines);
-    } else {
-      ctx.ui.setWidget(widgetKey, lines);
 			}
+			return lines;
 		};
 
-  // Smooth spinner animation at 100ms intervals
-  const spinnerTimer = setInterval(() => {
-    frameIndex = (frameIndex + 1) % SPINNER_FRAMES.length;
-    updateWidget();
-  }, 100);
+		ctx.ui.setWidget(widgetKey, (tui, t) => {
+			widgetTui = tui;
+			return {
+				render: (width?: number) => buildLines(t, width),
+				invalidate: () => widgetTui?.requestRender(),
+			};
+		});
+	}
 
-  // Initial display
-  updateWidget();
+	const requestRender = () => widgetTui?.requestRender();
+
+	// Spinner animation (sequential only — parallel uses a single batch timer)
+	let spinnerTimer: NodeJS.Timeout | undefined;
+	if (!isParallel) {
+		spinnerTimer = setInterval(() => {
+			frameIndex = (frameIndex + 1) % SPINNER_FRAMES.length;
+			requestRender();
+		}, 100);
+	}
 
 	// Use task-level timeout if set, otherwise fall back to config
 	const timeoutMs = task.timeoutMs ?? config.execution.timeoutMs;
 
-  // Pre-create session file path so events stream to disk (avoids 300+ MB in-memory accumulation)
-  const sessionsDir = path.join(ralpiDir, "sessions");
-  ensureDir(sessionsDir);
-  const sessionFilePath = path.join(sessionsDir, `${task.id}-${startMs}.txt`);
-
 	// Run task asynchronously via Pi SDK — event loop stays responsive
 	const output = await runAgentSession(
 		prompt,
@@ -142,32 +270,44 @@ export async function runTask(
 					name: event.toolName,
 					label,
 				});
-        updateWidget();
+				if (isParallel) {
+					const entry = parallelState!.get(task.id);
+					if (entry) {
+						entry.toolCalls.push({ name: event.toolName, label });
+					}
+					batchRender?.();
+				} else {
+					requestRender();
+				}
 			}
 		},
 		undefined, // no abort signal
-    sessionFilePath, // stream events to file
+		assignedModel ?? config.model,
+		config.thinkingLevel,
 	);
 
 	const durationMs = Date.now() - startMs;
 
 	// Clear progress widget and status after task finishes
-  clearInterval(spinnerTimer);
-  if (batcher) {
-    batcher.scheduleRemove(widgetKey);
+	if (spinnerTimer) clearInterval(spinnerTimer);
+	if (isParallel) {
+		const entry = parallelState!.get(task.id);
+		if (entry) {
+			entry.done = true;
+			entry.success = output.success;
+		}
+		batchRender?.();
 	} else {
 		ctx.ui.setWidget(widgetKey, undefined);
 	}
-  ctx.ui.setStatus("ralpi", undefined);
 
 	if (!output.success) {
-    sendChatMessage?.(`✗ ${taskHeader} — ${output.error}`);
-    ctx.ui.notify(`Task ${task.id} failed: ${output.error}`, "error");
+		// Failure reporting is handled by the caller (executeTask) to avoid
+		// duplicate messages when model failover or retry cycling is active.
 		return {
 			success: false,
 			error: output.error,
 			durationMs,
-      sessionFile: sessionFilePath, // events streamed to file for debugging
 		};
 	}
 
@@ -177,13 +317,10 @@ export async function runTask(
 	// Capture git commits made during this task
 	const { commitMessages, commitSummary } = captureGitCommits(projectDir);
 
-  // Session file already written by runAgentSession (events streamed to disk)
-  const sessionFile = sessionFilePath;
-
 	// Build output preview (first 500 chars of agent text)
 	const outputPreview =
 		agentText.length > 500
-      ? agentText.slice(0, 500) + "\n... (truncated, see session file)"
+			? agentText.slice(0, 500) + "\n... (truncated)"
 			: agentText;
 
 	// Extract reflection from agent output
@@ -199,7 +336,6 @@ export async function runTask(
 		durationMs,
 		toolUsage,
 		outputPreview,
-    sessionFile,
 		commitMessages,
 		commitSummary,
 	};
@@ -227,9 +363,43 @@ export async function executeBatch(
 		);
 	}
 
-  // Check if we should run parallel
+	// Set up model round-robin if configured.
+	// Config entries are "<provider>/<model>" strings — resolve via modelRegistry.
+	let roundRobin: ModelRoundRobin | null = null;
+	if (config.execution.models.length > 0) {
+		const resolvedModels: unknown[] = [];
+		for (const entry of config.execution.models) {
+			const slashIdx = entry.indexOf("/");
+			if (slashIdx === -1) {
+				ctx.ui.notify(
+					`ralpi config: skipping model "${entry}" — expected <provider>/<model> format`,
+					"warning",
+				);
+				continue;
+			}
+			const provider = entry.slice(0, slashIdx);
+			const modelId = entry.slice(slashIdx + 1);
+			const resolved = ctx.modelRegistry?.find(provider, modelId);
+			if (resolved) {
+				resolvedModels.push(resolved);
+			} else {
+				ctx.ui.notify(
+					`ralpi config: model "${entry}" not found in registry — skipping`,
+					"warning",
+				);
+			}
+		}
+		if (resolvedModels.length > 0) {
+			roundRobin = new ModelRoundRobin(resolvedModels);
+		}
+	}
+
+	// Check if we should run parallel.
+	// Use the parallel path whenever the user selected parallel mode,
+	// even for single-task batches produced by DAG dependency chains.
+	// Only sequential mode should inherit the parent session model.
 	const shouldParallel =
-    options?.parallel && tasks.length > 1 && config.execution.maxParallel > 0;
+		options?.parallel && tasks.length > 0 && config.execution.maxParallel > 0;
 
 	if (shouldParallel) {
 		await executeBatchParallel(
@@ -240,12 +410,14 @@ export async function executeBatch(
 			ctx,
 			sendChatMessage,
 			projectDir,
+			roundRobin,
 		);
 		return;
 	}
 
-  // Execute sequentially
+	// Execute sequentially (no round-robin — inherit parent model)
 	for (const task of tasks) {
+		try {
 			await executeTask(
 				task,
 				project,
@@ -255,6 +427,22 @@ export async function executeBatch(
 				sendChatMessage,
 				projectDir,
 			);
+		} catch (error) {
+			// Task failed — stop the batch. Dependent tasks are blocked by
+			// the DAG layer (getBlockedTasks) so they won't appear in this batch.
+
+			const errorMsg = error instanceof Error ? error.message : String(error);
+			progress.markFailed(task.id, errorMsg);
+			// Auto-update the PRD source file checkbox
+			try {
+				updateTaskInFile(project.sourcePath, task.id, "failed");
+			} catch {
+				// Best-effort
+			}
+			sendChatMessage?.(`✗ ${task.id} · ${task.title} — ${errorMsg}`);
+			ctx.ui.notify(`Task ${task.id} failed: ${errorMsg}`, "error");
+			break;
+		}
 	}
 }
 
@@ -269,15 +457,107 @@ async function executeBatchParallel(
 	ctx: ExtensionContext,
 	sendChatMessage?: SendChatMessage,
 	projectDir?: string,
+	roundRobin?: ModelRoundRobin | null,
 ): Promise<void> {
 	const maxParallel = config.execution.maxParallel;
-  const batcher = new WidgetBatcher(ctx);
-  const results: Array<{ task: Task; result: Promise<any> }> = [];
+	const sharedState: ParallelWidgetState = new Map();
 
-  for (const task of tasks) {
-    results.push({
-      task,
-      result: executeTask(
+	// Register a single batch widget that renders ALL parallel tasks in ID order.
+	const widgetKey = `ralpi-parallel-${Date.now()}`;
+	let widgetTui: { requestRender(): void } | null = null;
+
+	const buildBatchLines = (
+		t: typeof ctx.ui.theme,
+		width?: number,
+	): string[] => {
+		const effectiveWidth = width || 74;
+		const lines: string[] = [];
+		const sortedIds = Array.from(sharedState.keys()).sort();
+
+		for (const id of sortedIds) {
+			const entry = sharedState.get(id)!;
+			const frame = entry.done
+				? entry.success
+					? "✓"
+					: "✗"
+				: t.fg("accent", SPINNER_FRAMES[entry.frameIndex]);
+			lines.push(
+				truncateToWidth(`${frame} ${entry.taskHeader}`, effectiveWidth),
+			);
+
+			// Only show tool calls for in-progress tasks; completed/failed
+			// tasks already have their tool-call tree in the chat history message.
+			if (!entry.done && entry.toolCalls.length > 0) {
+				if (entry.toolCalls.length <= MAX_COLLAPSED) {
+					for (let i = 0; i < entry.toolCalls.length; i++) {
+						const tc = entry.toolCalls[i];
+						const isLast = i === entry.toolCalls.length - 1;
+						const branch = isLast ? "  └── " : "  ├── ";
+						const tag = t.fg("accent", `[${tc.name}]`);
+						lines.push(
+							truncateToWidth(`${branch}${tag} ${tc.label}`, effectiveWidth),
+						);
+					}
+				} else {
+					const shown = entry.toolCalls.slice(-MAX_COLLAPSED);
+					const remaining = entry.toolCalls.length - shown.length;
+					lines.push(
+						truncateToWidth(
+							t.fg("dim", `  ├── …${remaining} earlier`),
+							effectiveWidth,
+						),
+					);
+					for (let i = 0; i < shown.length; i++) {
+						const tc = shown[i];
+						const isLast = i === shown.length - 1;
+						const branch = isLast ? "  └── " : "  ├── ";
+						const tag = t.fg("accent", `[${tc.name}]`);
+						lines.push(
+							truncateToWidth(`${branch}${tag} ${tc.label}`, effectiveWidth),
+						);
+					}
+				}
+			}
+		}
+		return lines;
+	};
+
+	ctx.ui.setWidget(widgetKey, (tui, t) => {
+		widgetTui = tui;
+		return {
+			render: (width?: number) => buildBatchLines(t, width),
+			invalidate: () => widgetTui?.requestRender(),
+		};
+	});
+
+	// Batch-render trigger: re-render on spinner ticks AND content changes.
+	// Spinner animation requires requestRender() on every tick; without it,
+	// spinner frames advance in memory but the display never updates.
+	const requestBatchRender = () => widgetTui?.requestRender();
+
+	const spinnerTimer = setInterval(() => {
+		for (const entry of sharedState.values()) {
+			if (!entry.done) {
+				entry.frameIndex = (entry.frameIndex + 1) % SPINNER_FRAMES.length;
+			}
+		}
+		requestBatchRender();
+	}, 100);
+
+	// Semaphore-based concurrency control:
+	// Start up to maxParallel tasks immediately. When ANY task completes,
+	// start the next pending task. This ensures slots fill as soon as they
+	// open, instead of blocking on the oldest task (FIFO pattern).
+	const pending = [...tasks];
+	const running = new Set<Promise<void>>();
+
+	/** Start the next pending task if a slot is available. */
+	const kick = (): void => {
+		while (running.size < maxParallel && pending.length > 0) {
+			const task = pending.shift()!;
+			const assignedModel = roundRobin?.assign(task.id);
+
+			const p = executeTask(
 				task,
 				project,
 				config,
@@ -285,24 +565,51 @@ async function executeBatchParallel(
 				ctx,
 				sendChatMessage,
 				projectDir,
-        batcher,
-      ),
+				sharedState,
+				assignedModel,
+				roundRobin,
+				requestBatchRender,
+			)
+				.catch((error) => {
+					// Safety net: one task failure should never crash the batch.
+					// executeTask already marks failed and notifies, but catch as
+					// a last resort so the error doesn't propagate and crash pi.
+					roundRobin?.release(task.id);
+					requestBatchRender();
+					const errorMsg =
+						error instanceof Error ? error.message : String(error);
+					progress.markFailed(task.id, errorMsg);
+					// Auto-update the PRD source file checkbox
+					try {
+						updateTaskInFile(project.sourcePath, task.id, "failed");
+					} catch {
+						// Best-effort
+					}
+					sendChatMessage?.(`✗ ${task.id} · ${task.title} — ${errorMsg}`);
+					ctx.ui.notify(`Task ${task.id} failed: ${errorMsg}`, "error");
+				})
+				.finally(() => {
+					// Remove from running set and start next pending task
+					running.delete(p);
+					requestBatchRender();
+					kick();
 				});
 
-    // Limit concurrency
-    if (results.length >= maxParallel) {
-      const first = results.shift();
-      if (first) await first.result;
+			running.add(p);
 		}
+	};
+
+	// Kick off initial batch of tasks (up to maxParallel)
+	kick();
+
+	// Wait for all tasks to complete (kick() adds new promises to `running`
+	// when completed tasks free up slots, so we iterate until the set is empty).
+	while (running.size > 0) {
+		await Promise.race(running);
 	}
 
-  // Wait for remaining tasks
-  for (const { result } of results) {
-    await result;
-  }
-
-  // Flush and stop the batcher after all tasks complete
-  batcher.stop();
+	clearInterval(spinnerTimer);
+	ctx.ui.setWidget(widgetKey, undefined);
 }
 
 // ─── Execute Single Task with Retry ──────────────────────────────────────────
@@ -315,15 +622,39 @@ async function executeTask(
 	ctx: ExtensionContext,
 	sendChatMessage?: SendChatMessage,
 	projectDir: string = project.sourceDir,
-  batcher?: WidgetBatcher,
+	parallelState?: ParallelWidgetState,
+	assignedModel?: unknown,
+	roundRobin?: ModelRoundRobin | null,
+	batchRender?: () => void,
 ): Promise<void> {
 	const maxRetries = config.execution.maxRetries;
-  let retries = 0;
 
+	// Model failover: when a provider/API is down, cycle through available models.
+	// result.success === false always means an agent-session failure (API error,
+	// provider unreachable, etc.), not a task-work error.
+	const maxModelAttempts = roundRobin ? roundRobin.length : 1;
+	let modelAttempt = 0;
+	let currentModel: unknown = assignedModel ?? config.model;
+
+	while (modelAttempt < maxModelAttempts) {
+		// On subsequent model attempts, advance to the next model.
+		// Uses advance() instead of assign() so we don't get stuck on
+		// the same freed slot when the current model is down.
+		if (modelAttempt > 0 && roundRobin) {
+			currentModel = roundRobin.advance(task.id);
+		}
+
+		let retries = 0;
 		while (retries <= maxRetries) {
 			try {
 				// Mark as in progress
 				progress.markInProgress(task.id);
+				// Auto-update the PRD source file checkbox
+				try {
+					updateTaskInFile(project.sourcePath, task.id, "in_progress");
+				} catch {
+					// Best-effort: don't fail the task over a checkbox update
+				}
 
 				// Get dependency reflections
 				const depReflections = progress.getDependencyReflections(
@@ -339,10 +670,183 @@ async function executeTask(
 					ctx,
 					sendChatMessage,
 					projectDir,
-        batcher,
+					parallelState,
+					currentModel,
+					batchRender,
 				);
 
 				if (result.success) {
+					// ── Auto-Commit: Trigger follow-up agent session for uncommitted changes ──
+					let finalCommitMessages = result.commitMessages ?? [];
+					let finalCommitSummary = result.commitSummary ?? "";
+
+					try {
+						if (hasUncommittedChanges(projectDir)) {
+							const status = getGitStatusPorcelain(projectDir);
+							const diff = getGitDiff(projectDir);
+							const commitPrompt = [
+								`## Auto-Commit for Task ${task.id}: ${task.title}`,
+								"",
+								"The previous task is complete. There are uncommitted changes in the repository.",
+								"",
+								"Only commit changes you made while completing this task. Do not commit pre-existing changes, changes from other work, or files unrelated to this task.",
+								"Review the git status and diff below to identify which changes are from your work, and stage only those files.",
+								"",
+								"Stage only the files relevant to this task with `git add <files>`, then create a meaningful git commit.",
+								"Use a descriptive commit message and follow conventional commits format.",
+								"",
+								"### Current Changes (git status --porcelain)",
+								"```text",
+								status || "(no status output)",
+								"```",
+								"",
+								"### Current Tracked Diff (git diff)",
+								"```diff",
+								diff || "(no tracked diff output)",
+								"```",
+							].join("\n");
+
+							// ── Commit widget setup ──
+							const commitWidgetKey = `ralpi-commit-${task.id}`;
+							let commitFrameIndex = 0;
+							const commitToolCalls: ToolCallEntry[] = [];
+							let commitWidgetTui: { requestRender(): void } | null = null;
+
+							const commitHeader = `commit for ${task.id} · ${task.title}`;
+
+							const buildCommitLines = (
+								t: typeof ctx.ui.theme,
+								width?: number,
+							): string[] => {
+								const effectiveWidth = width || 74;
+								const frame = t.fg(
+									"accent",
+									SPINNER_FRAMES[commitFrameIndex % SPINNER_FRAMES.length],
+								);
+								const lines = [
+									truncateToWidth(`~ ${frame} ${commitHeader}`, effectiveWidth),
+								];
+
+								if (commitToolCalls.length > 0) {
+									if (commitToolCalls.length <= MAX_COLLAPSED) {
+										for (let i = 0; i < commitToolCalls.length; i++) {
+											const entry = commitToolCalls[i];
+											const isLast = i === commitToolCalls.length - 1;
+											const branch = isLast ? "  └── " : "  ├── ";
+											const tag = t.fg("accent", `[${entry.name}]`);
+											lines.push(
+												truncateToWidth(
+													`${branch}${tag} ${entry.label}`,
+													effectiveWidth,
+												),
+											);
+										}
+									} else {
+										const shown = commitToolCalls.slice(-MAX_COLLAPSED);
+										const remaining = commitToolCalls.length - shown.length;
+										lines.push(
+											truncateToWidth(
+												t.fg("dim", `  ├── …${remaining} earlier`),
+												effectiveWidth,
+											),
+										);
+										for (let i = 0; i < shown.length; i++) {
+											const entry = shown[i];
+											const isLast = i === shown.length - 1;
+											const branch = isLast ? "  └── " : "  ├── ";
+											const tag = t.fg("accent", `[${entry.name}]`);
+											lines.push(
+												truncateToWidth(
+													`${branch}${tag} ${entry.label}`,
+													effectiveWidth,
+												),
+											);
+										}
+									}
+								}
+								return lines;
+							};
+
+							ctx.ui.setWidget(commitWidgetKey, (tui, t) => {
+								commitWidgetTui = tui;
+								return {
+									render: (width?: number) => buildCommitLines(t, width),
+									invalidate: () => commitWidgetTui?.requestRender(),
+								};
+							});
+
+							const requestCommitRender = () =>
+								commitWidgetTui?.requestRender();
+
+							const commitSpinnerTimer = setInterval(() => {
+								commitFrameIndex =
+									(commitFrameIndex + 1) % SPINNER_FRAMES.length;
+								requestCommitRender();
+							}, 100);
+
+							// Use a short timeout for the commit session (60s should be enough)
+							const commitTimeout = Math.min(
+								60_000,
+								config.execution.timeoutMs,
+							);
+
+							let commitResult: Awaited<ReturnType<typeof runAgentSession>>;
+
+							try {
+								commitResult = await runAgentSession(
+									commitPrompt,
+									projectDir,
+									commitTimeout,
+									(event) => {
+										if (event.type === "tool_execution_start") {
+											const label = formatToolArg(event.toolName, event.args);
+											commitToolCalls.push({
+												name: event.toolName,
+												label,
+											});
+											requestCommitRender();
+										}
+									},
+									undefined,
+									currentModel,
+									config.thinkingLevel,
+								);
+							} finally {
+								clearInterval(commitSpinnerTimer);
+								ctx.ui.setWidget(commitWidgetKey, undefined);
+							}
+
+							if (commitResult.success) {
+								// Re-capture commits made during this follow-up session
+								const newCommits = captureGitCommits(projectDir);
+								if (newCommits.commitMessages.length > 0) {
+									finalCommitMessages = [
+										...finalCommitMessages,
+										...newCommits.commitMessages,
+									];
+									finalCommitSummary = finalCommitSummary
+										? `${finalCommitSummary}; ${newCommits.commitSummary}`
+										: newCommits.commitSummary;
+								}
+								sendChatMessage?.(`✓ commit for ${task.id} · ${task.title}`, {
+									toolCalls: commitToolCalls,
+								});
+							} else {
+								sendChatMessage?.(
+									`~ commit for ${task.id} · ${task.title} — follow-up commit session failed: ${commitResult.error}`,
+									{ toolCalls: commitToolCalls },
+								);
+							}
+						}
+					} catch (error) {
+						// Don't fail the task if auto-commit fails
+						sendChatMessage?.(
+							`~ commit for ${task.id} · ${task.title} — auto-commit error: ${
+								error instanceof Error ? error.message : String(error)
+							}`,
+						);
+					}
+
 					// Save reflection
 					if (result.reflection) {
 						saveReflectionToFile(projectDir, config, result.reflection);
@@ -354,20 +858,38 @@ async function executeTask(
 						result.durationMs,
 						result.reflection,
 						result.toolUsage,
-          result.sessionFile,
 						result.outputPreview,
-          result.commitMessages,
-          result.commitSummary,
+						finalCommitMessages,
+						finalCommitSummary,
 					);
+					// Auto-update the PRD source file checkbox
+					try {
+						updateTaskInFile(project.sourcePath, task.id, "completed");
+					} catch {
+						// Best-effort: don't fail the task over a checkbox update
+					}
+					roundRobin?.release(task.id);
 					return;
 				}
 
-      // Task failed, check if we should retry
+				// Agent session failed (provider error).
+				// If we have more models, cycle immediately — don't waste retries.
+				if (roundRobin && modelAttempt < maxModelAttempts - 1) {
+					// Don't release — advance() already handles the transition.
+					// release() would put the slot in freeSlots, then assign()
+					// would pick it right back up, getting stuck on the same model.
+					modelAttempt++;
+					sendChatMessage?.(
+						`~ ${task.id} · ${task.title} — trying model ${modelAttempt + 1}/${maxModelAttempts} (previous: ${result.error})`,
+					);
+					break; // exit retry loop, cycle to next model
+				}
+
+				// No more models — use normal retry logic
 				if (retries < maxRetries) {
 					retries = progress.incrementRetry(task.id);
-        ctx.ui.notify(
-          `Retrying task ${task.id} (${retries}/${maxRetries}): ${result.error}`,
-          "warning",
+					sendChatMessage?.(
+						`~ ${task.id} · ${task.title} — retrying (${retries}/${maxRetries}): ${result.error}`,
 					);
 
 					// Exponential backoff
@@ -376,14 +898,49 @@ async function executeTask(
 				} else {
 					// Max retries exceeded
 					progress.markFailed(task.id, result.error || "Unknown error");
-        throw new Error(`Task ${task.id} failed: ${result.error}`);
+					// Don't update PRD — retry exhaustion is transient, not terminal
+					sendChatMessage?.(`✗ ${task.id} · ${task.title} — ${result.error}`);
+					ctx.ui.notify(
+						`Task ${task.id} failed after ${maxRetries} retries: ${
+							result.error || "Unknown error"
+						}`,
+						"error",
+					);
+					return;
 				}
 			} catch (error) {
+				roundRobin?.release(task.id);
+				batchRender?.();
 				const errorMsg = error instanceof Error ? error.message : String(error);
 				progress.markFailed(task.id, errorMsg);
-      throw error;
+				// Auto-update the PRD source file checkbox
+				try {
+					updateTaskInFile(project.sourcePath, task.id, "failed");
+				} catch {
+					// Best-effort
+				}
+				sendChatMessage?.(`✗ ${task.id} · ${task.title} — ${errorMsg}`);
+				ctx.ui.notify(`Task ${task.id} failed: ${errorMsg}`, "error");
+				return;
 			}
 		}
+
+		// If we broke out (model cycling), continue the outer loop
+		modelAttempt++;
+	}
+
+	// All models exhausted — release the slot
+	roundRobin?.release(task.id);
+	batchRender?.();
+	progress.markFailed(task.id, "All configured models exhausted");
+	// Don't update PRD — model exhaustion is transient, not terminal
+	sendChatMessage?.(
+		`✗ ${task.id} · ${task.title} — all ${maxModelAttempts} models exhausted`,
+	);
+	ctx.ui.notify(
+		`Task ${task.id} failed: all configured models exhausted`,
+		"error",
+	);
 }
 
 // ─── Save Reflection to File ────────────────────────────────────────────────
@@ -407,6 +964,20 @@ function sleep(ms: number): Promise<void> {
 
 // ─── Tool Call Formatting ────────────────────────────────────────────────
 
+/**
+ * Strip control characters and newlines from a display label so it
+ * does not break TUI layout (tree branches, text width calculation).
+ */
+function sanitizeLabel(s: string): string {
+	// Replace newlines/carriage returns with spaces (multi-line commands
+	// must fit on a single tree-branch line), then strip ASCII control
+	// characters except \t (which is harmless) and keep printable chars.
+	return s
+		.replace(/\r?\n/g, " ")
+		.replace(/[\x00-\x08\x0B\x0C\x0E-\x1F\x7F]/g, "")
+		.trim();
+}
+
 /**
  * Format a tool call argument into a short label.
  */
@@ -414,21 +985,20 @@ function formatToolArg(name: string, args: unknown): string {
 	const a = args as Record<string, unknown>;
 	switch (name) {
 		case "bash":
-      return truncateMiddle(String(a.command ?? ""), 70);
+			return sanitizeLabel(truncateMiddle(String(a.command ?? ""), 70));
 		case "write":
 		case "read":
-      return truncateMiddle(String(a.path ?? ""), 60);
+			return sanitizeLabel(truncateMiddle(String(a.path ?? ""), 60));
 		case "edit":
-      return truncateMiddle(String(a.path ?? ""), 60);
+			return sanitizeLabel(truncateMiddle(String(a.path ?? ""), 60));
 		case "grep":
-      return `${a.pattern ?? "?"} — ${truncateMiddle(
-        String(a.path ?? ""),
-        40,
-      )}`;
+			return sanitizeLabel(
+				`${a.pattern ?? "?"} — ${truncateMiddle(String(a.path ?? ""), 40)}`,
+			);
 		case "find":
-      return `${a.path ?? "."} — ${a.glob ?? "*"}`;
+			return sanitizeLabel(`${a.path ?? "."} — ${a.glob ?? "*"}`);
 		case "ls":
-      return truncateMiddle(String(a.path ?? "."), 60);
+			return sanitizeLabel(truncateMiddle(String(a.path ?? "."), 60));
 		default:
 			return name;
 	}

src/parser.ts

@@ -1,6 +1,20 @@
 import * as fs from "node:fs";
 import * as path from "node:path";
-import type { Task, Project } from "./types";
+import type { Task, Project, ParallelGroup, Phase } from "./types";
+
+// Lazy-loaded yaml package
+let YAML_module: typeof import("yaml") | undefined;
+function loadYaml(): typeof import("yaml") {
+	if (YAML_module) return YAML_module;
+	try {
+		YAML_module = require("yaml");
+	} catch {
+		throw new Error(
+			"YAML parsing requires the 'yaml' package. Run: npm install yaml",
+		);
+	}
+	return YAML_module!;
+}
 
 // ─── Main Entry ──────────────────────────────────────────────────────────────
 
@@ -8,6 +22,7 @@ import type { Task, Project } from "./types";
  * Parse a task file (markdown or YAML) into a Project structure.
  * Supports:
  * - Fio README format (numbered tasks with dependency graph)
+ * - Phased format (## Phase N — Title sections with tasks and dependencies)
  * - Simple checkbox format (- [ ] task)
  * - YAML format (tasks: [...])
  */
@@ -22,7 +37,7 @@ export function parseTaskFile(filePath: string): Project {
 	}
 
 	// Markdown: detect format
-	if (hasDependenciesSection(content)) {
+	if (hasDependenciesSection(content) || hasPhaseHeadings(content)) {
 		return parseFioFormat(content, absolutePath, dir);
 	}
 	return parseSimpleCheckbox(content, absolutePath, dir);
@@ -30,8 +45,35 @@ export function parseTaskFile(filePath: string): Project {
 
 // ─── Fio Format Parser ───────────────────────────────────────────────────────
 
+/** Match both markdown heading (## Dependencies) and plain heading (Dependencies). */
+const DEP_HEADING_RE = /^(?:##\s+)?Dependencies\s*$/m;
+/** Match both markdown heading (## Tasks) and plain heading (Tasks). */
+const TASK_HEADING_RE = /^(?:##\s+)?Tasks\s*$/m;
+/** Match other markdown headings (## Something). */
+const ANY_MD_HEADING_RE = /^##\s/;
+/** Match phase headings: ## Phase 1 — Push-to-Talk MVP */
+const PHASE_HEADING_RE = /^\s*##\s+Phase\s+(\d+)\s*[—–:-]\s*(.+)$/i;
+/** Detect plain phase headings too: Phase 1 — Title (no ##) */
+const PHASE_HEADING_PLAIN_RE = /^Phase\s+(\d+)\s*[—–:-]\s*(.+)$/i;
+/**
+ * Detect a plain (non-markdown) section heading like "Exit criteria".
+ * A plain heading must:
+ *   - Start with a letter
+ *   - Contain only letters and spaces
+ *   - Have no colons (avoids matching "Objective:" and "Status legend:")
+ *   - Not be a task/dep line (doesn't start with "-")
+ */
+function isPlainSectionHeader(line: string): boolean {
+	const trimmed = line.trim();
+	return trimmed.length > 0 && /^[A-Za-z][A-Za-z\s]*$/.test(trimmed);
+}
+
 function hasDependenciesSection(content: string): boolean {
-	return /^##\s+Dependencies\s*$/m.test(content);
+	return DEP_HEADING_RE.test(content);
+}
+
+function hasPhaseHeadings(content: string): boolean {
+	return PHASE_HEADING_RE.test(content) || PHASE_HEADING_PLAIN_RE.test(content);
 }
 
 function parseFioFormat(
@@ -42,24 +84,57 @@ function parseFioFormat(
 	const lines = content.split("\n");
 	const tasks: Task[] = [];
 	const dependencies: Record<string, string[]> = {};
+	const parallelGroups: ParallelGroup[] = [];
+	const phases: Phase[] = [];
+	let currentPhase: number | null = null;
+	let currentPhaseTitle = "";
 	let inTasks = false;
 	let inDeps = false;
 
 	for (const line of lines) {
-		if (/^##\s+Tasks\s*$/m.test(line)) {
+		// Check for phase headings first
+		const phaseMatch =
+			line.match(PHASE_HEADING_RE) || line.match(PHASE_HEADING_PLAIN_RE);
+		if (phaseMatch) {
+			// Save previous phase if exists
+			if (currentPhase !== null) {
+				const phaseTaskIds = tasks
+					.filter((t) => t.phase === currentPhase)
+					.map((t) => t.id);
+				if (phaseTaskIds.length > 0) {
+					phases.push({
+						number: currentPhase,
+						title: currentPhaseTitle,
+						taskIds: phaseTaskIds,
+					});
+				}
+			}
+			// Start new phase
+			currentPhase = parseInt(phaseMatch[1], 10);
+			currentPhaseTitle = phaseMatch[2].trim();
 			inTasks = true;
 			inDeps = false;
 			continue;
 		}
-		if (/^##\s+Dependencies\s*$/m.test(line)) {
+
+		if (TASK_HEADING_RE.test(line)) {
+			inTasks = true;
+			inDeps = false;
+			continue;
+		}
+		if (DEP_HEADING_RE.test(line)) {
 			inTasks = false;
 			inDeps = true;
 			continue;
 		}
+		// Reset state on any other section heading — both ##-style and plain
+		// BUT NOT phase headings (already handled above)
 		if (
-			/^##\s/.test(line) &&
-			!/^##\s+Tasks/.test(line) &&
-			!/^##\s+Dependencies/.test(line)
+			(ANY_MD_HEADING_RE.test(line) || isPlainSectionHeader(line)) &&
+			!TASK_HEADING_RE.test(line) &&
+			!DEP_HEADING_RE.test(line) &&
+			!PHASE_HEADING_RE.test(line) &&
+			!PHASE_HEADING_PLAIN_RE.test(line)
 		) {
 			inTasks = false;
 			inDeps = false;
@@ -67,15 +142,17 @@ function parseFioFormat(
 		}
 
 		if (inTasks) {
-			// Match all tasks on a line (supports compact single-line formats)
+			// Match all tasks on a line (supports compact single-line formats).
+			// ID is digits optionally followed by a single lowercase letter
+			// (e.g. "01", "02b", "10c") — see normalizeTaskId for the shape.
 			const taskPattern =
-				/-+\s+\[(.)\]\s+(\d+)\s+[—–:-]\s+(.+?)(?:\s+(?=-+\s+\[)|\s*→\s*`([^`]+)`|$)/g;
+				/-+\s+\[(.)\]\s+(\d+[a-z]?)\s+[—–:-]\s+(.+?)(?:\s+(?=-+\s+\[)|\s*→\s*`([^`]+)`|$)/g;
 			let match: RegExpExecArray | null;
 			while ((match = taskPattern.exec(line)) !== null) {
 				const [, status, id, title, file] = match;
 				const timeoutMs = parseTimeoutFromLine(line);
 				tasks.push({
-					id: `0${id}`,
+					id: normalizeTaskId(id),
 					title: title.trim(),
 					description: undefined,
 					file: file || undefined,
@@ -83,71 +160,229 @@ function parseFioFormat(
 					dependencies: [],
 					timeoutMs,
 					index: tasks.length,
+					phase: currentPhase ?? undefined,
 				});
 			}
 		}
 
 		if (inDeps) {
-			// Format 2: Arrow notation with multiple targets
-			// "01 -> 02,03,06 (description)" means 02, 03, 06 depend on 01
+			// Arrow notation (supports both -> and unicode \u2192)
+			// "01 -> 02,03,06" means 02, 03, 06 depend on 01
+			// "02 \u2192 08" — single arrow with unicode
+			// "03 \u2192 04 \u2192 05" — chained: 04 depends on 03, 05 depends on 04
+			// "05, 07, 08 \u2192 13" — multi-prereq: 13 depends on 05, 07, 08
 			// Supports optional markdown list prefix: "- 01 -> 02,03,06"
-			const arrowMatch = line.match(
-				/^(?:\s*[-*]\s+)?(\d+)\s*->\s*([\d,\s]+?)(?:\s*\(|$)/,
-			);
-			if (arrowMatch) {
-				const [, from, targets] = arrowMatch;
-				const fromId = `0${from}`;
-				const targetIds = targets
+			const hasArrow = /->/.test(line) || /\u2192/.test(line);
+			if (hasArrow) {
+				// Strip optional list prefix and parenthetical description
+				const cleaned = line
+					.replace(/^(\s*[-*]\s+)?/, "")
+					.replace(/\s*\(.*\)\s*$/, "");
+
+				// Split on arrows to get segments
+				const segments = cleaned
+					.split(/->|\u2192/)
+					.map((s) => s.trim())
+					.filter(Boolean);
+
+				if (segments.length >= 2) {
+					for (let i = 0; i < segments.length - 1; i++) {
+						// Left segment: source(s) (comma-separated)
+						const fromIds = segments[i]
 							.split(",")
 							.map((t) => t.trim())
-					.filter((t) => t)
-					.map((t) => `0${t}`);
+							.filter((t) => /^\d+[a-z]?$/.test(t))
+							.map((t) => normalizeTaskId(t));
 
-				// Each target depends on the source
-				for (const toId of targetIds) {
+						// Right segment: target(s) (comma-separated)
+						const toIds = segments[i + 1]
+							.split(",")
+							.map((t) => t.trim())
+							.filter((t) => /^\d+[a-z]?$/.test(t))
+							.map((t) => normalizeTaskId(t));
+
+						for (const toId of toIds) {
 							if (!dependencies[toId]) dependencies[toId] = [];
+							for (const fromId of fromIds) {
+								if (!dependencies[toId].includes(fromId)) {
 									dependencies[toId].push(fromId);
 								}
 							}
+						}
+					}
+				}
+			}
 
 			// Format 1: Natural language "X depends on A, B, C"
 			// Supports optional markdown list prefix: "- 13 depends on 17, 18, 19"
+			// Also handles "also depends on": "- 08 also depends on 05, 06"
+			// The dep list char class includes lowercase letters so lettered IDs
+			// (e.g. "02b") don't truncate the capture. Per-id validation is
+			// done by the filter below, so trailing prose can't leak in.
 			const dependsMatch = line.match(
-				/^(?:\s*[-*]\s+)?(\d+)\s+depends\s+on\s+([\d,\s]+)/i,
+				/^(?:\s*[-*]\s+)?(\d+[a-z]?)\s+(?:also\s+)?depends\s+on\s+([\d,\s a-z]+)/i,
 			);
 			if (dependsMatch) {
 				const [, taskId, depsList] = dependsMatch;
-				const taskIdPadded = `0${taskId}`;
+				const taskIdPadded = normalizeTaskId(taskId);
 				const depIds = depsList
 					.split(",")
 					.map((t) => t.trim())
-					.filter((t) => t)
-					.map((t) => `0${t}`);
+					.filter((t) => /^\d+[a-z]?$/.test(t))
+					.map((t) => normalizeTaskId(t));
 
 				if (!dependencies[taskIdPadded]) dependencies[taskIdPadded] = [];
-				dependencies[taskIdPadded].push(...depIds);
+				for (const depId of depIds) {
+					if (!dependencies[taskIdPadded].includes(depId)) {
+						dependencies[taskIdPadded].push(depId);
+					}
+				}
 			}
 
 			// Parse meta blocks for task configuration (timeout, etc.)
 			const metaMatch = line.match(
-				/^0?(\d+)\s+\[timeout\]\s*=?\s*(\d+)(?:m|min|s|ms)?/i,
+				/^0?(\d+[a-z]?)\s+\[timeout\]\s*=?\s*(\d+)(?:m|min|s|ms)?/i,
 			);
 			if (metaMatch) {
 				const [, taskId, value, unit] = metaMatch;
-				const task = tasks.find((t) => t.id === `0${taskId}`);
+				const task = tasks.find((t) => t.id === normalizeTaskId(taskId));
 				if (task) {
 					task.timeoutMs = parseTimeoutValue(Number(value), unit);
 				}
 			}
+
+			// Format 2: "X, Y, Z can be done in parallel (label)"
+			// "- 01, 02, 03, 04 can be done in parallel (Play Store prep)"
+			const parallelMatch = line.match(
+				/^(?:\s*[-*]\s+)?((?:0?\d+[a-z]?\s*,\s*)*0?\d+[a-z]?)\s+can\s+be\s+done\s+in\s+parallel(?:\s+\(([^)]+)\))?$/i,
+			);
+			if (parallelMatch) {
+				const [, idsStr, label] = parallelMatch;
+				const taskIds = idsStr
+					.split(",")
+					.map((t) => t.trim())
+					.filter((t) => /^\d+[a-z]?$/.test(t))
+					.map((t) => normalizeTaskId(t));
+
+				if (taskIds.length > 0) {
+					parallelGroups.push({
+						index: parallelGroups.length,
+						label: label ? label.trim() : undefined,
+						taskIds,
+					});
+				}
+			}
+
+			// Format 3: "A must be done before B, C" or "A, B must be done before C"
+			// "- 21 must be done before 22, 23, 24 (backend integration foundation)"
+			// "- 02, 03 must be done before 04"
+			const mustBeforeMatch = line.match(
+				/^(?:\s*[-*]\s+)?((?:0?\d+[a-z]?\s*,\s*)*0?\d+[a-z]?)\s+must\s+be\s+done\s+before\s+((?:0?\d+[a-z]?\s*,\s*)*0?\d+[a-z]?)(?:\s+\(([^)]+)\))?$/i,
+			);
+			if (mustBeforeMatch) {
+				const [, fromIdsStr, toIdsStr] = mustBeforeMatch;
+				const fromIds = fromIdsStr
+					.split(",")
+					.map((t) => t.trim())
+					.filter((t) => /^\d+[a-z]?$/.test(t))
+					.map((t) => normalizeTaskId(t));
+				const toIds = toIdsStr
+					.split(",")
+					.map((t) => t.trim())
+					.filter((t) => /^\d+[a-z]?$/.test(t))
+					.map((t) => normalizeTaskId(t));
+
+				// Each "to" task depends on ALL "from" tasks
+				for (const toId of toIds) {
+					if (!dependencies[toId]) dependencies[toId] = [];
+					for (const fromId of fromIds) {
+						if (!dependencies[toId].includes(fromId)) {
+							dependencies[toId].push(fromId);
+						}
+					}
 				}
 			}
 
-	// Extract exit criteria
+			// Format 4: "X, Y, Z depend on A" or "X depends on A, B, C"
+			// "- 22, 23, 24 depend on 21"
+			// "- 05, 06 depend on 02, 03, 04"
+			// "- 08 also depends on 05, 06"  ("also" is ignored)
+			// Strip optional "also" before matching
+			const cleanedLine = line.replace(/\balso\b/i, "");
+			const dependOnMatch = cleanedLine.match(
+				/^(?:\s*[-*]\s+)?((?:0?\d+[a-z]?\s*,\s*)*0?\d+[a-z]?)\s+depend(?:s)?\s+on\s+((?:0?\d+[a-z]?\s*,\s*)*0?\d+[a-z]?)(?:\s+\(([^)]+)\))?$/i,
+			);
+			if (dependOnMatch) {
+				const [, fromIdsStr, toIdsStr] = dependOnMatch;
+				const fromIds = fromIdsStr
+					.split(",")
+					.map((t) => t.trim())
+					.filter((t) => /^\d+[a-z]?$/.test(t))
+					.map((t) => normalizeTaskId(t));
+				const toIds = toIdsStr
+					.split(",")
+					.map((t) => t.trim())
+					.filter((t) => /^\d+[a-z]?$/.test(t))
+					.map((t) => normalizeTaskId(t));
+
+				// Each "from" task depends on ALL "to" tasks
+				for (const fromId of fromIds) {
+					if (!dependencies[fromId]) dependencies[fromId] = [];
+					for (const toId of toIds) {
+						if (!dependencies[fromId].includes(toId)) {
+							dependencies[fromId].push(toId);
+						}
+					}
+				}
+			}
+		}
+	}
+
+	// Save final phase if we were in one
+	if (currentPhase !== null) {
+		const phaseTaskIds = tasks
+			.filter((t) => t.phase === currentPhase)
+			.map((t) => t.id);
+		if (phaseTaskIds.length > 0) {
+			phases.push({
+				number: currentPhase,
+				title: currentPhaseTitle,
+				taskIds: phaseTaskIds,
+			});
+		}
+	}
+
+	// Add implicit phase-boundary dependencies
+	// First task of each phase (except phase 1) depends on last task of previous phase
+	if (phases.length > 1) {
+		for (let i = 1; i < phases.length; i++) {
+			const prevPhase = phases[i - 1];
+			const currPhase = phases[i];
+			if (prevPhase.taskIds.length === 0 || currPhase.taskIds.length === 0)
+				continue;
+
+			const lastTaskOfPrevPhase =
+				prevPhase.taskIds[prevPhase.taskIds.length - 1];
+			const firstTaskOfCurrPhase = currPhase.taskIds[0];
+
+			// Add dependency if not already present
+			if (!dependencies[firstTaskOfCurrPhase]) {
+				dependencies[firstTaskOfCurrPhase] = [];
+			}
+			if (!dependencies[firstTaskOfCurrPhase].includes(lastTaskOfPrevPhase)) {
+				dependencies[firstTaskOfCurrPhase].push(lastTaskOfPrevPhase);
+			}
+		}
+	}
+
+	// Extract exit criteria — detect both ## Exit Criteria and plain Exit criteria
 	const exitCriteria: string[] = [];
-	const exitIdx = lines.findIndex((l) => /^##\s+Exit\s+Criteria/i.test(l));
+	const exitCriteriaRe = /^(?:##\s+)?Exit\s+Criteria/i;
+	const exitIdx = lines.findIndex((l) => exitCriteriaRe.test(l));
 	if (exitIdx >= 0) {
 		for (let i = exitIdx + 1; i < lines.length; i++) {
-			if (/^##\s/.test(lines[i])) break;
+			// Stop at any new section heading (##-style or plain)
+			if (/^##\s/.test(lines[i]) || isPlainSectionHeader(lines[i])) break;
 			const m = lines[i].match(/^-\s+(.+)$/);
 			if (m) exitCriteria.push(m[1].trim());
 		}
@@ -164,9 +399,21 @@ function parseFioFormat(
 		}
 	}
 
+	// Apply parallelGroup to tasks
+	for (const group of parallelGroups) {
+		for (const taskId of group.taskIds) {
+			const task = tasks.find((t) => t.id === taskId);
+			if (task) {
+				task.parallelGroup = group.index;
+			}
+		}
+	}
+
 	return {
 		tasks,
 		dependencies,
+		parallelGroups: parallelGroups.length > 0 ? parallelGroups : undefined,
+		phases: phases.length > 0 ? phases : undefined,
 		sourcePath,
 		sourceDir,
 		exitCriteria,
@@ -210,16 +457,7 @@ function parseYaml(
 	sourcePath: string,
 	sourceDir: string,
 ): Project {
-	// Lazy-load yaml (may not be installed)
-	let YAML: typeof import("yaml");
-	try {
-		YAML = require("yaml");
-	} catch {
-		throw new Error(
-			"YAML parsing requires the 'yaml' package. Run: npm install yaml",
-		);
-	}
-
+	const YAML = loadYaml();
 	const doc = YAML.parse(content);
 	const tasks: Task[] = [];
 
@@ -263,35 +501,119 @@ export function readTaskSpec(taskDir: string, taskFile: string): string {
 // ─── Task File Updater ───────────────────────────────────────────────────────
 
 /**
- * Update task status in the source markdown file
+ * Update task status in the source file (markdown or YAML).
+ *
+ * Handles three formats:
+ * 1. Fio numbered format: `- [ ] 01 – Title` — matches by task number in the file
+ * 2. Simple checkbox: `- [ ] Title` — matches by checkbox position (index)
+ * 3. YAML: uses `yaml` library to parse, update, and stringify
  */
 export function updateTaskInFile(
 	filePath: string,
 	taskId: string,
 	status: Task["status"],
 ): void {
-	let content = fs.readFileSync(filePath, "utf-8");
-	const char = statusToChar(status);
+	const ext = path.extname(filePath).toLowerCase();
 
-	// Try Fio numbered format first
-	const fioPattern = new RegExp(
-		`(^-\\s+\\[)(.)(\\]\\s+${escapeRegex(taskId)}\\s*[—–-])`,
-		"m",
-	);
-	if (fioPattern.test(content)) {
-		content = content.replace(fioPattern, `$1${char}$3`);
-		fs.writeFileSync(filePath, content, "utf-8");
+	// Handle YAML format
+	if (ext === ".yaml" || ext === ".yml") {
+		updateTaskInYaml(filePath, taskId, status);
 		return;
 	}
 
-	// Try simple checkbox format
-	const simplePattern = new RegExp(
-		`(-\\s+\\[)(.)(\\]\\s+${escapeRegex(taskId)})`,
+	let content = fs.readFileSync(filePath, "utf-8");
+	const char = statusToChar(status);
+
+	// Strategy 1: Fio numbered format — match by explicit task ID in the file.
+	// For pure-digit IDs, also try the parsed numeric form (parity with the
+	// pre-lettered behavior). Lettered IDs ("02b", "02c") only have one valid
+	// form — the parseInt fallback would silently drop the letter suffix and
+	// create false-positive partial matches, so we skip it for them.
+	const idPatterns = new Set([escapeRegex(taskId)]);
+	if (!taskId.startsWith("0") && /^\d+$/.test(taskId)) {
+		const rawId = parseInt(taskId, 10).toString();
+		idPatterns.add(escapeRegex(rawId));
+	}
+
+	for (const idPattern of idPatterns) {
+		const fioRegex = new RegExp(
+			`(^-\\s+\\[)(.)(\\]\\s+${idPattern}\\s*[—–:-])`,
 			"m",
 		);
-	if (simplePattern.test(content)) {
-		content = content.replace(simplePattern, `$1${char}$3`);
+		const match = content.match(fioRegex);
+		if (match) {
+			content = content.replace(fioRegex, `$1${char}$3`);
 			fs.writeFileSync(filePath, content, "utf-8");
+			return;
+		}
+	}
+
+	// Strategy 2: Simple checkbox by position (task IDs are zero-padded indices)
+	const targetIndex = parseInt(taskId, 10);
+	if (!isNaN(targetIndex)) {
+		const lines = content.split("\n");
+		let checkboxIdx = 0;
+		for (let i = 0; i < lines.length; i++) {
+			const m = lines[i].match(/^(\s*-+\s+\[)(.)(\].*)$/);
+			if (m) {
+				if (checkboxIdx === targetIndex) {
+					lines[i] = m[1] + char + m[3];
+					fs.writeFileSync(filePath, lines.join("\n"), "utf-8");
+					return;
+				}
+				checkboxIdx++;
+			}
+		}
+	}
+}
+
+/**
+ * Update task status in a YAML task file using the yaml library's
+ * Document API, which preserves comments and formatting.
+ *
+ * Matches by explicit `id` field first, then falls back to
+ * position-based matching (for files without explicit IDs).
+ */
+function updateTaskInYaml(
+	filePath: string,
+	taskId: string,
+	status: Task["status"],
+): void {
+	const YAML = loadYaml();
+	const content = fs.readFileSync(filePath, "utf-8");
+	const doc = YAML.parseDocument(content);
+	const tasks = doc.get("tasks");
+	if (!tasks || !YAML.isSeq(tasks)) return;
+
+	// Build alternate ID forms for matching. For lettered IDs ("02b"), the
+	// verbatim form is the only valid pattern — parseInt would drop the suffix.
+	const idVariants: string[] = [taskId];
+	if (/^\d+$/.test(taskId)) {
+		idVariants.push(parseInt(taskId, 10).toString());
+	}
+
+	// Strategy 1: Match by explicit id field
+	for (const item of tasks.items) {
+		if (!YAML.isMap(item)) continue;
+		const idVal = item.get("id");
+		if (idVal === undefined || idVal === null) continue;
+		const idStr = String(idVal);
+		if (idVariants.includes(idStr)) {
+			item.set("status", status);
+			fs.writeFileSync(filePath, String(doc), "utf-8");
+			return;
+		}
+	}
+
+	// Strategy 2: Fall back to position-based matching
+	// (for YAML files without explicit id fields)
+	const targetIndex = parseInt(taskId, 10);
+	if (!isNaN(targetIndex) && targetIndex < tasks.items.length) {
+		const item = tasks.items[targetIndex];
+		if (YAML.isMap(item)) {
+			item.set("status", status);
+			fs.writeFileSync(filePath, String(doc), "utf-8");
+		}
 	}
 }
 
@@ -392,6 +714,28 @@ function parseTimeoutFromMeta(
 	return undefined;
 }
 
+/**
+ * Normalize a task ID: zero-pad the digit portion to 2 chars, preserve any
+ * single lowercase letter suffix. Idempotent on already-normalized IDs.
+ *
+ *   "1"   → "01"
+ *   "2"   → "02"
+ *   "2b"  → "02b"
+ *   "02b" → "02b"
+ *   "10"  → "10"
+ *   "10b" → "10b"
+ *
+ * Pass-through for IDs that don't match the expected shape (defensive — the
+ * upstream regexes restrict matches, but a stray value should not be silently
+ * re-shaped).
+ */
+function normalizeTaskId(id: string): string {
+	const match = id.match(/^(\d+)([a-z])?$/);
+	if (!match) return id;
+	const [, digits, letter] = match;
+	return digits.padStart(2, "0") + (letter ?? "");
+}
+
 function charToStatus(char: string): Task["status"] {
 	switch (char) {
 		case " ":

src/progress.ts

@@ -171,7 +171,6 @@ export class ProgressTracker {
 		durationMs: number,
 		reflection?: Reflection,
 		toolUsage?: ToolUsage,
-    sessionFile?: string,
 		outputPreview?: string,
 		commitMessages?: string[],
 		commitSummary?: string,
@@ -183,7 +182,6 @@ export class ProgressTracker {
 		prd.tasks[taskId].durationMs = durationMs;
 		if (reflection) prd.tasks[taskId].reflection = reflection;
 		if (toolUsage) prd.tasks[taskId].toolUsage = toolUsage;
-    if (sessionFile) prd.tasks[taskId].sessionFile = sessionFile;
 		if (outputPreview) prd.tasks[taskId].outputPreview = outputPreview;
 		if (commitMessages) prd.tasks[taskId].commitMessages = commitMessages;
 		if (commitSummary) prd.tasks[taskId].commitSummary = commitSummary;
@@ -213,6 +211,14 @@ export class ProgressTracker {
 			.map(([id]) => id);
 	}
 
+	/** Get IDs of all failed tasks */
+	getFailedTaskIds(): string[] {
+		const prd = this.getPRD();
+		return Object.entries(prd.tasks)
+			.filter(([, info]) => info.status === "failed")
+			.map(([id]) => id);
+	}
+
 	/** Get all reflections from completed tasks */
 	getAllReflections(): Reflection[] {
 		const prd = this.getPRD();

src/types.ts

@@ -27,6 +27,26 @@ export interface Task {
 	timeoutMs?: number;
 	/** Original index in task list for deterministic ordering */
 	index?: number;
+	/** Phase number this task belongs to (1-indexed, from ## Phase N headings) */
+	phase?: number;
+}
+
+export interface ParallelGroup {
+	/** Group index (0-based, determines execution order) */
+	index: number;
+	/** Human-readable label for the group (e.g. "Play Store prep") */
+	label?: string;
+	/** Task IDs in this group — all can run concurrently */
+	taskIds: string[];
+}
+
+export interface Phase {
+	/** Phase number (1-indexed, matches the heading number) */
+	number: number;
+	/** Phase title (e.g. "Push-to-Talk MVP") */
+	title: string;
+	/** Task IDs in this phase, in order */
+	taskIds: string[];
 }
 
 export interface Project {
@@ -36,6 +56,10 @@ export interface Project {
 	tasks: Task[];
 	/** Explicit dependency map: taskId → [dependency taskIds] */
 	dependencies: Record<string, string[]>;
+	/** Explicit parallel groups from "can be done in parallel" declarations */
+	parallelGroups?: ParallelGroup[];
+	/** Phased sections from ## Phase N headings (in order) */
+	phases?: Phase[];
 	/** Exit criteria (from README ## Exit Criteria section) */
 	exitCriteria?: string[];
 	/** Path to the source task file */
@@ -97,8 +121,6 @@ export interface TaskProgressInfo {
 	error?: string;
 	/** Tool usage counts from parsed subprocess output */
 	toolUsage?: ToolUsage;
-	/** Path to session output file */
-	sessionFile?: string;
 	/** Truncated output preview for expanded view */
 	outputPreview?: string;
 	/** Git commit messages from task execution */
@@ -153,6 +175,8 @@ export interface RalpiConfig {
 		timeoutMs: number;
 		/** Maximum parallel tasks (0 = unlimited) */
 		maxParallel: number;
+		/** Round-robin model list for parallel tasks (empty = inherit parent model) */
+		models: string[];
 	};
 	prompts: {
 		/** Additional context injected into every task prompt */
@@ -160,6 +184,10 @@ export interface RalpiConfig {
 		/** Custom prompt suffix for reflection extraction */
 		reflectionPrompt: string;
 	};
+	/** Parent session model to inherit in child agent sessions */
+	model?: unknown;
+	/** Parent session thinking level to inherit in child agent sessions */
+	thinkingLevel?: unknown;
 }
 
 export const DEFAULT_CONFIG: RalpiConfig = {
@@ -168,10 +196,11 @@ export const DEFAULT_CONFIG: RalpiConfig = {
 		reflectionsDir: ".ralpi/reflections",
 	},
 	execution: {
-		maxRetries: 3,
-		retryDelayMs: 5000,
-		timeoutMs: 30 * 60 * 1000, // 30 minutes
+		maxRetries: 0,
+		retryDelayMs: 0,
+		timeoutMs: 0, // 0 = inherit Pi's own defaults (no ralpi-level timeout)
 		maxParallel: 3,
+		models: [],
 	},
 	prompts: {
 		projectContext: "",

src/utils.ts

@@ -34,6 +34,78 @@ export function writeFileSafe(filePath: string, content: string): void {
 	fs.writeFileSync(filePath, content, "utf-8");
 }
 
+// ─── Loop-Active State ──────────────────────────────────────────────────────
+
+/**
+ * State persisted to disk when a ralpi execution loop is active.
+ * Used to re-instantiate widgets after a session reload.
+ */
+export interface LoopActiveState {
+	taskFile: string;
+	mode: "parallel" | "sequential";
+	startedAt: string;
+	taskIds: string[];
+	prdKey: string;
+}
+
+/**
+ * Path (relative to projectDir) where the loop-active marker is stored.
+ */
+const LOOP_ACTIVE_FILE = ".ralpi/loop-active.json";
+
+/**
+ * Write the loop-active marker, indicating an execution loop is running.
+ */
+export function writeLoopActive(
+	projectDir: string,
+	state: LoopActiveState,
+): void {
+	writeFileSafe(
+		path.join(projectDir, LOOP_ACTIVE_FILE),
+		JSON.stringify(state, null, 2),
+	);
+}
+
+/**
+ * Read the loop-active marker, if present.
+ */
+export function readLoopActive(projectDir: string): LoopActiveState | null {
+	const filePath = path.join(projectDir, LOOP_ACTIVE_FILE);
+	try {
+		const raw = fs.readFileSync(filePath, "utf-8");
+		return JSON.parse(raw) as LoopActiveState;
+	} catch {
+		return null;
+	}
+}
+
+/**
+ * Delete the loop-active marker.
+ */
+export function deleteLoopActive(projectDir: string): void {
+	const filePath = path.join(projectDir, LOOP_ACTIVE_FILE);
+	try {
+		fs.unlinkSync(filePath);
+	} catch {
+		// Ignore if already gone
+	}
+}
+
+/**
+ * Discover the project directory by walking up to find `.ralpi/`.
+ */
+export function findRalpiDir(startDir: string): string | null {
+	let current = path.resolve(startDir);
+	const root = path.parse(current).root;
+	while (current !== root) {
+		if (fs.existsSync(path.join(current, ".ralpi"))) {
+			return current;
+		}
+		current = path.dirname(current);
+	}
+	return null;
+}
+
 // ─── Async Agent Session ────────────────────────────────────────────────────
 
 // ─── Progress Discovery ─────────────────────────────────────────────────────
@@ -82,32 +154,35 @@ export function findProgressFile(
 
 // ─── Config ──────────────────────────────────────────────────────────────────
 
-function parseSimpleYaml(content: string): Record<string, any> {
+/** Try to use the `yaml` package (real dependency in package.json).
+ *  Falls back to a flat key:value parser when unavailable. */
+const parseSimpleYaml: (content: string) => Record<string, any> = (() => {
+	try {
+		// eslint-disable-next-line @typescript-eslint/no-var-requires
+		const { parse } = require("yaml");
+		return (content: string) => parse(content) ?? {};
+	} catch {
+		return (content: string) => {
 			const result: Record<string, any> = {};
-	const lines = content.split("\n");
-
-	for (const line of lines) {
+			for (const line of content.split("\n")) {
 				const trimmed = line.trim();
 				if (!trimmed || trimmed.startsWith("#")) continue;
-
-		const match = trimmed.match(/^([^:]+):\s*(.+)$/);
+				const match = trimmed.match(/^([^:]+):\s*(.*)$/);
 				if (match) {
-			const key = match[1].trim();
-			let value: string | boolean | number = match[2].trim();
-
-			// Parse booleans
-			if (value === "true") value = true;
-			else if (value === "false") value = false;
-			// Parse numbers
-			else if (/^\d+$/.test(value)) value = parseInt(value, 10);
-			else if (/^\d+\.\d+$/.test(value)) value = parseFloat(value);
-
-			result[key] = value;
+					const value = match[2].trim();
+					if (value === "true") result[match[1].trim()] = true;
+					else if (value === "false") result[match[1].trim()] = false;
+					else if (/^\d+$/.test(value))
+						result[match[1].trim()] = parseInt(value, 10);
+					else if (/^\d+\.\d+$/.test(value))
+						result[match[1].trim()] = parseFloat(value);
+					else result[match[1].trim()] = value;
 				}
 			}
-
 			return result;
-}
+		};
+	}
+})();
 
 /**
  * Deep merge configuration objects
@@ -129,25 +204,44 @@ function mergeConfig(
 	return result as RalpiConfig;
 }
 
+/** Path to the global ralpi config under the user's Pi home directory. */
+const GLOBAL_CONFIG_PATH = path.join(
+	process.env.HOME || "/tmp",
+	".pi",
+	"ralpi",
+	"config.yaml",
+);
+
 /**
- * Load configuration from .ralpi/config.yaml or return defaults
+ * Load and merge config from global and project sources.
+ *
+ * Precedence (highest wins):
+ *   1. Project-level: `<projectDir>/.ralpi/config.yaml`
+ *   2. Global: `~/.pi/ralpi/config.yaml`
+ *   3. `DEFAULT_CONFIG` in `src/types.ts`
  */
 export function loadConfig(projectDir: string): RalpiConfig {
-	const configPath = path.join(projectDir, ".ralpi", "config.yaml");
+	// Start with defaults
+	const merged: RalpiConfig = { ...DEFAULT_CONFIG };
 
-	// Return defaults silently when config file does not exist
-	if (!fs.existsSync(configPath)) {
-		return { ...DEFAULT_CONFIG };
-	}
+	// Layer 1: global config (~/.pi/ralpi/config.yaml)
+	tryLoadConfigFile(GLOBAL_CONFIG_PATH, merged);
 
+	// Layer 2: project config (.ralpi/config.yaml) — overrides global
+	tryLoadConfigFile(path.join(projectDir, ".ralpi", "config.yaml"), merged);
+
+	return merged;
+
+	/** Attempt to load a single config file and merge into `acc` in place. */
+	function tryLoadConfigFile(filePath: string, acc: RalpiConfig): void {
+		if (!fs.existsSync(filePath)) return;
 		try {
-		const content = fs.readFileSync(configPath, "utf-8");
-		// Simple YAML parsing (key: value format)
-		const config = parseSimpleYaml(content);
-		return mergeConfig(DEFAULT_CONFIG, config);
+			const content = fs.readFileSync(filePath, "utf-8");
+			const parsed = parseSimpleYaml(content);
+			Object.assign(acc, mergeConfig(acc, parsed));
 		} catch {
-		// Malformed config — fall back to defaults silently
-		return { ...DEFAULT_CONFIG };
+			// Malformed config — skip silently
+		}
 	}
 }
 
@@ -338,7 +432,8 @@ export async function runAgentSession(
 	timeoutMs: number,
 	onEvent?: (event: AgentSessionEvent) => void,
 	signal?: AbortSignal,
-	sessionFile?: string,
+	model?: unknown,
+	thinkingLevel?: unknown,
 ): Promise<{
 	success: boolean;
 	text: string;
@@ -354,17 +449,13 @@ export async function runAgentSession(
 		bash: 0,
 		other: 0,
 	};
-	// Stream events to file instead of accumulating in memory.
-	// Accumulating caused "Invalid string length" crashes when
-	// JSON.stringify(output.events, null, 2) produced 300+ MB strings.
-	const eventStream = sessionFile
-		? fs.createWriteStream(sessionFile, { flags: "a" })
-		: null;
-
-	// Wire timeout via abort signal
-	const timeoutHandle = setTimeout(() => {
+	// Wire timeout via abort signal (only when set; 0 means inherit Pi's defaults)
+	let timeoutHandle: NodeJS.Timeout | null = null;
+	if (timeoutMs > 0) {
+		timeoutHandle = setTimeout(() => {
 			if (sessionRef?.session) sessionRef.session.agent.abort();
 		}, timeoutMs);
+	}
 
 	const sessionRef: {
 		session?: Awaited<ReturnType<typeof createAgentSession>>["session"];
@@ -387,6 +478,8 @@ export async function runAgentSession(
 			sessionManager: SessionManager.inMemory(),
 			resourceLoader: loader,
 			tools: ["read", "bash", "edit", "write", "grep", "find", "ls"],
+			model: model as any,
+			thinkingLevel: thinkingLevel as any,
 		});
 		sessionRef.session = result.session;
 
@@ -399,10 +492,6 @@ export async function runAgentSession(
 		let stopReason: string | undefined;
 
 		const unsubscribe = result.session.subscribe((event) => {
-			// Stream event to file (avoids accumulating 300+ MB in memory)
-			if (eventStream) {
-				eventStream.write(JSON.stringify(event) + "\n");
-			}
 			onEvent?.(event);
 
 			if (event.type === "message_end") {
@@ -437,12 +526,7 @@ export async function runAgentSession(
 		unsubscribe();
 		result.session.dispose();
 		signal?.removeEventListener("abort", abortHandler);
-		clearTimeout(timeoutHandle);
-
-		// Flush and close the event stream before returning
-		if (eventStream) {
-			await new Promise<void>((resolve) => eventStream.end(resolve));
-		}
+		if (timeoutHandle) clearTimeout(timeoutHandle);
 
 		if (errorMessage && !finalText) {
 			return {
@@ -460,19 +544,16 @@ export async function runAgentSession(
 			text: finalText.trim(),
 			toolUsage,
 			stopReason,
-			events: [], // streamed to file
+			events: [],
 		};
 	} catch (error) {
-		clearTimeout(timeoutHandle);
-		if (eventStream && !eventStream.destroyed) {
-			eventStream.end();
-		}
+		if (timeoutHandle) clearTimeout(timeoutHandle);
 		return {
 			success: false,
 			text: "",
 			error: error instanceof Error ? error.message : String(error),
 			toolUsage,
-			events: [], // streamed to file
+			events: [],
 		};
 	} finally {
 		sessionRef.session?.dispose();
@@ -498,6 +579,53 @@ function extractAssistantText(content: unknown): string {
 
 // ─── Git Commit Capture ──────────────────────────────────────────────────────
 
+/**
+ * Check if there are any uncommitted changes in the git repository.
+ */
+export function hasUncommittedChanges(projectDir: string): boolean {
+	const { execSync } = require("node:child_process");
+	try {
+		const output = execSync("git status --porcelain", {
+			cwd: projectDir,
+			encoding: "utf-8",
+		}).trim();
+		return output.length > 0;
+	} catch {
+		return false;
+	}
+}
+
+/**
+ * Get the current git status in porcelain format.
+ * Includes untracked files, which `git diff` alone would miss.
+ */
+export function getGitStatusPorcelain(projectDir: string): string {
+	const { execSync } = require("node:child_process");
+	try {
+		return execSync("git status --porcelain", {
+			cwd: projectDir,
+			encoding: "utf-8",
+		}).trim();
+	} catch {
+		return "";
+	}
+}
+
+/**
+ * Get the current git diff for tracked uncommitted changes.
+ */
+export function getGitDiff(projectDir: string): string {
+	const { execSync } = require("node:child_process");
+	try {
+		return execSync("git diff", {
+			cwd: projectDir,
+			encoding: "utf-8",
+		}).trim();
+	} catch {
+		return "";
+	}
+}
+
 /**
  * Capture recent git commits made during task execution
  * Returns commit messages and a summary string

tests/dag-construction.test.ts

@@ -0,0 +1,674 @@
+/// <reference types="bun-types" />
+import { describe, it, expect } from "bun:test";
+import type { Project, Task } from "../src/types";
+import {
+	buildExecutionPlan,
+	buildSequentialPlan,
+	getBlockedTasks,
+	detectCycles,
+	getCriticalPath,
+	formatDependencyChain,
+	formatExecutionPlan,
+} from "../src/dag";
+
+// ─── Helpers ─────────────────────────────────────────────────────────────────
+
+function makeProject(overrides?: Partial<Project>): Project {
+	return {
+		tasks: [],
+		dependencies: {},
+		sourcePath: "/tmp/test.md",
+		sourceDir: "/tmp",
+		...overrides,
+	};
+}
+
+function task(
+	id: string,
+	dependencies: string[] = [],
+	status: Task["status"] = "pending",
+	parallelGroup?: number,
+): Task {
+	return { id, title: `Task ${id}`, status, dependencies, parallelGroup };
+}
+
+function tasksFrom(...args: Task[]): Task[] {
+	return args;
+}
+
+// ─── Basic DAG Construction ──────────────────────────────────────────────────
+
+describe("buildExecutionPlan (Kahn's algorithm)", () => {
+	it("handles empty task list", () => {
+		const project = makeProject({ tasks: [] });
+		const plan = buildExecutionPlan(project, new Set());
+		expect(plan.batches).toEqual([]);
+		expect(plan.totalTasks).toBe(0);
+	});
+
+	it("puts all root tasks in batch 0", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02"), task("03")),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		expect(plan.batches).toHaveLength(1);
+		expect(plan.batches[0].tasks.map((t) => t.id).sort()).toEqual([
+			"01",
+			"02",
+			"03",
+		]);
+	});
+
+	it("builds correct linear dependency chain", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["02"]),
+				task("04", ["03"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		expect(plan.batches).toHaveLength(4);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[1].tasks.map((t) => t.id)).toEqual(["02"]);
+		expect(plan.batches[2].tasks.map((t) => t.id)).toEqual(["03"]);
+		expect(plan.batches[3].tasks.map((t) => t.id)).toEqual(["04"]);
+	});
+
+	it("groups parallelizable tasks in the same batch", () => {
+		// Diamond: 01 -> 02, 03 -> 04
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04", ["02", "03"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		// Batch 0: [01], Batch 1: [02, 03], Batch 2: [04]
+		expect(plan.batches).toHaveLength(3);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[1].tasks.map((t) => t.id).sort()).toEqual(["02", "03"]);
+		expect(plan.batches[2].tasks.map((t) => t.id)).toEqual(["04"]);
+	});
+
+	it("assigns correct batchIndex values", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04", ["02", "03"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		expect(plan.batches[0].batchIndex).toBe(0);
+		expect(plan.batches[1].batchIndex).toBe(1);
+		expect(plan.batches[2].batchIndex).toBe(2);
+	});
+
+	it("skips completed tasks and includes them in skippedTasks", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01", [], "completed"),
+				task("02", ["01"]),
+				task("03", ["02"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set(["01"]));
+		expect(plan.totalTasks).toBe(2);
+		expect(plan.skippedTasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches).toHaveLength(2);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["02"]);
+		expect(plan.batches[1].tasks.map((t) => t.id)).toEqual(["03"]);
+	});
+
+	it("throws on dependency cycle", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01", ["03"]),
+				task("02", ["01"]),
+				task("03", ["02"]),
+			),
+		});
+		expect(() => buildExecutionPlan(project, new Set())).toThrow(
+			/dependency cycle/i,
+		);
+	});
+
+	it("blocks tasks that depend on failed tasks", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["02"]),
+				task("04", ["03"]),
+			),
+		});
+		const plan = buildExecutionPlan(
+			project,
+			new Set(),
+			undefined,
+			new Set(["01"]),
+		);
+		// 01 is excluded from pending (failed). 02, 03, 04 are pending but
+		// transitively blocked — they don't appear in batches.
+		expect(plan.skippedTasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.totalTasks).toBe(3); // 02, 03, 04 are pending but blocked
+		expect(plan.batches).toHaveLength(0);
+	});
+
+	it("blocks immediate dependents when task fails", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04"), // independent
+			),
+		});
+		const plan = buildExecutionPlan(
+			project,
+			new Set(),
+			undefined,
+			new Set(["01"]),
+		);
+		// 01 is excluded from pending (failed). 02, 03 are pending but blocked
+		// (depend on 01). 04 is independent and ready.
+		expect(plan.skippedTasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["04"]);
+	});
+});
+
+// ─── Complex DAGs ───────────────────────────────────────────────────────────
+
+describe("Complex DAG batching", () => {
+	it("builds the OAuth PRD example correctly", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["02"]),
+				task("04", ["01"]),
+				task("05", ["03", "04"]),
+				task("06", ["03", "04"]),
+				task("07", ["03"]),
+				task("08", ["05", "06", "07"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		// Expected batches: [01], [02,04], [03], [05,06,07], [08]
+		expect(plan.batches).toHaveLength(5);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[1].tasks.map((t) => t.id).sort()).toEqual(["02", "04"]);
+		expect(plan.batches[2].tasks.map((t) => t.id)).toEqual(["03"]);
+		expect(plan.batches[3].tasks.map((t) => t.id).sort()).toEqual([
+			"05",
+			"06",
+			"07",
+		]);
+		expect(plan.batches[4].tasks.map((t) => t.id)).toEqual(["08"]);
+	});
+
+	it("builds the Design Token PRD example correctly", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04", ["02", "03"]),
+				task("05", ["04", "01"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		// Expected batches: [01], [02,03], [04], [05]
+		expect(plan.batches).toHaveLength(4);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[1].tasks.map((t) => t.id).sort()).toEqual(["02", "03"]);
+		expect(plan.batches[2].tasks.map((t) => t.id)).toEqual(["04"]);
+		expect(plan.batches[3].tasks.map((t) => t.id)).toEqual(["05"]);
+	});
+
+	it("handles a 3-tier diamond", () => {
+		//      01
+		//     /  \
+		//    02  03
+		//   / \ / \
+		//  04 05 06
+		//   \ | /
+		//    07
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04", ["02"]),
+				task("05", ["02", "03"]),
+				task("06", ["03"]),
+				task("07", ["04", "05", "06"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		expect(plan.batches).toHaveLength(4);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[1].tasks.map((t) => t.id).sort()).toEqual(["02", "03"]);
+		expect(plan.batches[2].tasks.map((t) => t.id).sort()).toEqual([
+			"04",
+			"05",
+			"06",
+		]);
+		expect(plan.batches[3].tasks.map((t) => t.id)).toEqual(["07"]);
+	});
+
+	it("handles a wide fan-out with delayed convergence", () => {
+		// 01 -> 02,03,04,05,06
+		// 02,03 -> 07
+		// 04,05 -> 08
+		// 06 -> 09
+		// 07,08,09 -> 10
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04", ["01"]),
+				task("05", ["01"]),
+				task("06", ["01"]),
+				task("07", ["02", "03"]),
+				task("08", ["04", "05"]),
+				task("09", ["06"]),
+				task("10", ["07", "08", "09"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		expect(plan.batches).toHaveLength(4);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[1].tasks.map((t) => t.id).sort()).toEqual([
+			"02",
+			"03",
+			"04",
+			"05",
+			"06",
+		]);
+		expect(plan.batches[2].tasks.map((t) => t.id).sort()).toEqual([
+			"07",
+			"08",
+			"09",
+		]);
+		expect(plan.batches[3].tasks.map((t) => t.id)).toEqual(["10"]);
+	});
+
+	it("handles multiple independent subgraphs", () => {
+		// Two completely independent chains:
+		// Chain A: 01 -> 02 -> 03
+		// Chain B: 04 -> 05
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["02"]),
+				task("04"),
+				task("05", ["04"]),
+			),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		// Batch 0: [01, 04] (both roots)
+		// Batch 1: [02, 05]
+		// Batch 2: [03]
+		expect(plan.batches[0].tasks.map((t) => t.id).sort()).toEqual(["01", "04"]);
+		expect(plan.batches[1].tasks.map((t) => t.id).sort()).toEqual(["02", "05"]);
+		expect(plan.batches[2].tasks.map((t) => t.id)).toEqual(["03"]);
+	});
+
+	it("batches tasks respecting fan-in convergence", () => {
+		// 01 -> 03, 02 -> 03 (03 depends on both 01 AND 02)
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02"), task("03", ["01", "02"])),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		expect(plan.batches[0].tasks.map((t) => t.id).sort()).toEqual(["01", "02"]);
+		expect(plan.batches[1].tasks.map((t) => t.id)).toEqual(["03"]);
+	});
+});
+
+// ─── Sequential Plan ─────────────────────────────────────────────────────────
+
+describe("buildSequentialPlan", () => {
+	it("puts each task in its own batch", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02", ["01"]), task("03", ["01"])),
+		});
+		const plan = buildSequentialPlan(project, new Set());
+		expect(plan.batches).toHaveLength(3);
+		plan.batches.forEach((b, i) => {
+			expect(b.tasks).toHaveLength(1);
+			expect(b.batchIndex).toBe(i);
+		});
+	});
+
+	it("skips completed tasks and blocks transitively failed tasks", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["02"]),
+				task("04"),
+			),
+		});
+		const plan = buildSequentialPlan(project, new Set(["01"]), new Set(["01"]));
+		// 01 failed => 02, 03 blocked. 04 independent, runs.
+		expect(plan.skippedTasks.map((t) => t.id).sort()).toEqual([
+			"01",
+			"02",
+			"03",
+		]);
+		expect(plan.totalTasks).toBe(3);
+	});
+
+	it("maintains task order in sequential batches", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02", ["01"]), task("03", ["01"])),
+		});
+		const plan = buildSequentialPlan(project, new Set());
+		expect(plan.batches.map((b) => b.tasks[0].id)).toEqual(["01", "02", "03"]);
+	});
+});
+
+// ─── getBlockedTasks ─────────────────────────────────────────────────────────
+
+describe("getBlockedTasks", () => {
+	it("returns direct dependents of failed tasks", () => {
+		const pending = tasksFrom(task("01"), task("02", ["01"]), task("03"));
+		const blocked = getBlockedTasks(pending, new Set(["01"]));
+		expect([...blocked]).toEqual(["02"]);
+	});
+
+	it("returns transitive dependents (chain reaction)", () => {
+		const pending = tasksFrom(
+			task("01"),
+			task("02", ["01"]),
+			task("03", ["02"]),
+			task("04", ["03"]),
+		);
+		const blocked = getBlockedTasks(pending, new Set(["01"]));
+		expect([...blocked].sort()).toEqual(["02", "03", "04"]);
+	});
+
+	it("does not affect tasks in separate subgraphs", () => {
+		const pending = tasksFrom(
+			task("01"),
+			task("02", ["01"]),
+			task("10"),
+			task("11", ["10"]),
+		);
+		const blocked = getBlockedTasks(pending, new Set(["01"]));
+		expect([...blocked].sort()).toEqual(["02"]);
+	});
+
+	it("returns empty set when no tasks depend on failed tasks", () => {
+		const pending = tasksFrom(task("01"), task("02"), task("03"));
+		const blocked = getBlockedTasks(pending, new Set(["99"]));
+		expect(blocked.size).toBe(0);
+	});
+});
+
+// ─── detectCycles ────────────────────────────────────────────────────────────
+
+describe("detectCycles", () => {
+	it("returns empty for acyclic graph", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02", ["01"]), task("03", ["02"])),
+		});
+		expect(detectCycles(project)).toEqual([]);
+	});
+
+	it("detects a 3-node cycle", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01", ["03"]),
+				task("02", ["01"]),
+				task("03", ["02"]),
+			),
+		});
+		const cycles = detectCycles(project);
+		expect(cycles.length).toBeGreaterThan(0);
+	});
+
+	it("detects a self-loop", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01", ["01"])),
+		});
+		const cycles = detectCycles(project);
+		expect(cycles.length).toBeGreaterThan(0);
+	});
+
+	it("detects cycle in disconnected subgraph", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"), // isolated
+				task("02", ["03"]),
+				task("03", ["02"]), // cycle
+			),
+		});
+		const cycles = detectCycles(project);
+		expect(cycles.length).toBeGreaterThan(0);
+	});
+
+	it("returns empty for graph with only diamond patterns", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04", ["02", "03"]),
+			),
+		});
+		expect(detectCycles(project)).toEqual([]);
+	});
+});
+
+// ─── getCriticalPath ─────────────────────────────────────────────────────────
+
+describe("getCriticalPath", () => {
+	it("returns the longest path through the DAG", () => {
+		// 01 -> 02 -> 03 -> 04  (long = 4)
+		// 01 -> 05 -> 04        (short = 3)
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["02"]),
+				task("04", ["03", "05"]),
+				task("05", ["01"]),
+			),
+		});
+		const path = getCriticalPath(project);
+		expect(path.length).toBe(4);
+		expect(path[0].id).toBe("01");
+		expect(path[path.length - 1].id).toBe("04");
+	});
+
+	it("returns single-node path for roots", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02"), task("03")),
+		});
+		const path = getCriticalPath(project);
+		expect(path.length).toBe(1);
+	});
+
+	it("handles complex branching by picking the longest chain", () => {
+		// 01 -> 02 -> 03 -> 04 -> 05  (long = 5)
+		// 01 -> 06 -> 05               (short = 3)
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["02"]),
+				task("04", ["03"]),
+				task("05", ["04", "06"]),
+				task("06", ["01"]),
+			),
+		});
+		const path = getCriticalPath(project);
+		// Should pick 01 -> 02 -> 03 -> 04 -> 05 (length 5)
+		expect(path.length).toBe(5);
+		expect(path.map((t) => t.id)).toEqual(["01", "02", "03", "04", "05"]);
+	});
+});
+
+// ─── formatDependencyChain ───────────────────────────────────────────────────
+
+describe("formatDependencyChain", () => {
+	it("renders a simple tree", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02", ["01"])),
+		});
+		const formatted = formatDependencyChain(project);
+		expect(formatted).toContain("01");
+		expect(formatted).toContain("02");
+	});
+
+	it("mentions root tasks", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02")),
+		});
+		const formatted = formatDependencyChain(project);
+		expect(formatted).toMatch(/01.*root|root.*01/i);
+	});
+
+	it("handles empty task list", () => {
+		const project = makeProject({ tasks: [] });
+		const formatted = formatDependencyChain(project);
+		expect(formatted).toContain("no tasks");
+	});
+
+	it("shows orphan tasks when dependencies reference non-existent IDs", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01", ["99"])),
+		});
+		const formatted = formatDependencyChain(project);
+		expect(formatted).toMatch(/orphan|unreached/i);
+	});
+});
+
+// ─── formatExecutionPlan ─────────────────────────────────────────────────────
+
+describe("formatExecutionPlan", () => {
+	it("displays task counts and batches", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02", ["01"])),
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		const formatted = formatExecutionPlan(plan);
+		expect(formatted).toContain("Total tasks");
+		expect(formatted).toContain("Batches");
+		expect(formatted).toContain("01");
+		expect(formatted).toContain("02");
+	});
+
+	it("shows skipped tasks", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01", [], "completed"), task("02", ["01"])),
+		});
+		const plan = buildExecutionPlan(project, new Set(["01"]));
+		const formatted = formatExecutionPlan(plan);
+		expect(formatted).toContain("completed");
+	});
+
+	it("shows parallel group annotations when provided", () => {
+		const project = makeProject({
+			tasks: tasksFrom(task("01"), task("02", ["01"]), task("03", ["01"])),
+			parallelGroups: [{ index: 0, label: "UI sprint", taskIds: ["02", "03"] }],
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		const formatted = formatExecutionPlan(plan, project.parallelGroups);
+		expect(formatted).toContain("UI sprint");
+	});
+});
+
+// ─── Group-Aware Batching ────────────────────────────────────────────────────
+
+describe("Parallel group batching", () => {
+	it("builds batches when parallel groups are defined", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04", ["02", "03"]),
+			),
+			parallelGroups: [
+				{ index: 0, label: "Frontend", taskIds: ["01", "02", "03", "04"] },
+			],
+		});
+		// Should route through buildGroupAwareBatches
+		const plan = buildExecutionPlan(project, new Set());
+		expect(plan.batches.length).toBeGreaterThan(0);
+	});
+
+	it("respects intra-group dependencies in parallel groups", () => {
+		// Tasks: 01 -> 02, 01 -> 03, 02 -> 04, 03 -> 04
+		// With parallel groups, there are no cross-group dependencies by definition.
+		// Intra-group deps are respected by Kahn's algorithm.
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01"),
+				task("02", ["01"]),
+				task("03", ["01"]),
+				task("04", ["02", "03"]),
+			),
+			parallelGroups: [
+				{ index: 0, label: "All", taskIds: ["01", "02", "03", "04"] },
+			],
+		});
+		const plan = buildExecutionPlan(project, new Set());
+		// Batch 0: [01], Batch 1: [02, 03], Batch 2: [04]
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[1].tasks.map((t) => t.id).sort()).toEqual(["02", "03"]);
+		expect(plan.batches[2].tasks.map((t) => t.id)).toEqual(["04"]);
+	});
+});
+
+// ─── Real-World Scenario: Resume with completed tasks ───────────────────────
+
+describe("Real-world resume scenarios", () => {
+	it("buildExecutionPlan correctly excludes file-based [x] completions", () => {
+		// Design Token PRD resume: 01,02,03 [x] in file, 04 [~], 05 [ ]
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01", [], "completed"),
+				task("02", ["01"], "completed"),
+				task("03", ["01"], "completed"),
+				task("04", ["02", "03"], "in_progress"),
+				task("05", ["04", "01"], "pending"),
+			),
+		});
+		// buildCompletedSet in index.ts produces {01, 02, 03} from file + progress
+		// This simulates what happens after buildCompletedSet is called
+		const completedFromFile = new Set(
+			project.tasks.filter((t) => t.status === "completed").map((t) => t.id),
+		);
+		const plan = buildExecutionPlan(project, completedFromFile);
+
+		// Only 04 and 05 should be pending
+		expect(plan.totalTasks).toBe(2);
+		expect(plan.batches).toHaveLength(2);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["04"]);
+		expect(plan.batches[1].tasks.map((t) => t.id)).toEqual(["05"]);
+	});
+
+	it("skipsTasks includes both progress-completed and file-completed tasks", () => {
+		const project = makeProject({
+			tasks: tasksFrom(
+				task("01", [], "completed"),
+				task("02", ["01"], "pending"),
+			),
+		});
+		// Simulate: 01 completed in file AND in progress
+		const plan = buildExecutionPlan(project, new Set(["01"]));
+		expect(plan.skippedTasks.map((t) => t.id)).toEqual(["01"]);
+		expect(plan.batches[0].tasks.map((t) => t.id)).toEqual(["02"]);
+	});
+});

tests/helpers.ts

@@ -0,0 +1,30 @@
+import * as fs from "node:fs";
+import * as path from "node:path";
+import * as os from "node:os";
+
+// ─── Helpers ────────────────────────────────────────────────────────────────
+
+/**
+ * Create a temporary directory for test files.
+ * Returns the path and a cleanup function.
+ */
+export function tempDir(): { dir: string; cleanup: () => void } {
+	const dir = fs.mkdtempSync(path.join(os.tmpdir(), "ralpi-test-"));
+	return {
+		dir,
+		cleanup: () => fs.rmSync(dir, { recursive: true, force: true }),
+	};
+}
+
+/**
+ * Write content to a temp markdown file and return its path.
+ */
+export function writeTaskFile(
+	dir: string,
+	name: string,
+	content: string,
+): string {
+	const filePath = path.join(dir, name);
+	fs.writeFileSync(filePath, content, "utf-8");
+	return filePath;
+}

tests/parser-phased.test.ts

@@ -0,0 +1,521 @@
+/**
+ * Tests for phased task format parsing
+ * Covers: phase detection, task parsing, phase boundaries, implicit dependencies
+ */
+
+import { describe, test, expect } from "bun:test";
+import { parseTaskFile } from "../src/parser";
+import type { Task } from "../src/types";
+import { tempDir, writeTaskFile } from "./helpers";
+
+/** Parse a task file from an inline template literal. */
+function parse(content: string) {
+	const { dir, cleanup } = tempDir();
+	try {
+		const filePath = writeTaskFile(dir, "README.md", content);
+		return { project: parseTaskFile(filePath), cleanup };
+	} catch (e) {
+		cleanup();
+		throw e;
+	}
+}
+
+describe("Phased task format", () => {
+	describe("Phase detection", () => {
+		test("detects phased format with markdown headings", () => {
+			const content = `# Voice Conversation
+
+## Phase 1 - MVP
+- [ ] 01 - Build voice pipeline
+- [ ] 02 - Add audio playback
+
+## Phase 2 - Streaming
+- [ ] 03 - WebSocket channel
+- [ ] 04 - Streaming STT
+
+## Dependencies
+- 02 depends on 01
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.phases).toBeDefined();
+				expect(project.phases?.length).toBe(2);
+				expect(project.phases?.[0].number).toBe(1);
+				expect(project.phases?.[0].title).toBe("MVP");
+				expect(project.phases?.[1].number).toBe(2);
+				expect(project.phases?.[1].title).toBe("Streaming");
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("detects phased format with plain headings", () => {
+			const content = `# Voice Conversation
+
+Phase 1 - MVP
+- [ ] 01 - Build voice pipeline
+- [ ] 02 - Add audio playback
+
+Phase 2 - Streaming
+- [ ] 03 - WebSocket channel
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.phases).toBeDefined();
+				expect(project.phases?.length).toBe(2);
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("supports various separators in phase headings", () => {
+			const variants = [
+				"## Phase 1 - MVP",
+				"## Phase 1 - MVP",
+				"## Phase 1 - MVP",
+				"## Phase 1: MVP",
+				"## Phase 1 -  MVP", // multiple spaces
+			];
+
+			for (const heading of variants) {
+				const content = `# Test
+
+${heading}
+- [ ] 01 - Task
+
+## Dependencies
+`;
+				const { project, cleanup } = parse(content);
+				try {
+					expect(project.phases).toBeDefined();
+					expect(project.phases?.length).toBe(1);
+				} finally {
+					cleanup();
+				}
+			}
+		});
+
+		test("handles phase headings with extra whitespace", () => {
+			const content = `# Test
+
+  ## Phase 1 - MVP
+- [ ] 01 - Task
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.phases).toBeDefined();
+				expect(project.phases?.length).toBe(1);
+			} finally {
+				cleanup();
+			}
+		});
+	});
+
+	describe("Task parsing within phases", () => {
+		test("assigns phase number to tasks", () => {
+			const content = `# Test
+
+## Phase 1 - MVP
+- [ ] 01 - Task A
+- [ ] 02 - Task B
+
+## Phase 2 - Enhancement
+- [ ] 03 - Task C
+- [ ] 04 - Task D
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.tasks[0].id).toBe("01");
+				expect(project.tasks[0].phase).toBe(1);
+				expect(project.tasks[1].id).toBe("02");
+				expect(project.tasks[1].phase).toBe(1);
+				expect(project.tasks[2].id).toBe("03");
+				expect(project.tasks[2].phase).toBe(2);
+				expect(project.tasks[3].id).toBe("04");
+				expect(project.tasks[3].phase).toBe(2);
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("tracks task IDs in each phase", () => {
+			const content = `# Test
+
+## Phase 1 - Foundation
+- [ ] 01 - Setup
+- [ ] 02 - Config
+
+## Phase 2 - Implementation
+- [ ] 03 - Feature A
+- [ ] 04 - Feature B
+- [ ] 05 - Feature C
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.phases?.[0].taskIds).toEqual(["01", "02"]);
+				expect(project.phases?.[1].taskIds).toEqual(["03", "04", "05"]);
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("handles tasks with different statuses in phases", () => {
+			const content = `# Test
+
+## Phase 1 - MVP
+- [x] 01 - Done task
+- [ ] 02 - Pending task
+- [~] 03 - In progress
+
+## Phase 2 - Next
+- [ ] 04 - Future task
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.tasks[0].status).toBe("completed");
+				expect(project.tasks[1].status).toBe("pending");
+				expect(project.tasks[2].status).toBe("in_progress");
+				expect(project.tasks[3].status).toBe("pending");
+
+				expect(project.phases?.[0].taskIds).toEqual(["01", "02", "03"]);
+				expect(project.phases?.[1].taskIds).toEqual(["04"]);
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("handles empty phases", () => {
+			const content = `# Test
+
+## Phase 1 - Empty
+
+## Phase 2 - Has tasks
+- [ ] 01 - Task
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.phases?.length).toBe(1);
+				expect(project.phases?.[0].number).toBe(2);
+				expect(project.phases?.[0].taskIds).toEqual(["01"]);
+			} finally {
+				cleanup();
+			}
+		});
+	});
+
+	describe("Implicit phase-boundary dependencies", () => {
+		test("adds dependency from first task of phase 2 to last task of phase 1", () => {
+			const content = `# Test
+
+## Phase 1 - MVP
+- [ ] 01 - Setup
+- [ ] 02 - Build
+
+## Phase 2 - Enhancement
+- [ ] 03 - Feature
+- [ ] 04 - Test
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				// Task 03 should depend on task 02 (implicit phase boundary)
+				expect(
+					project.tasks.find((t: Task) => t.id === "03")?.dependencies,
+				).toContain("02");
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("adds dependencies across multiple phases", () => {
+			const content = `# Test
+
+## Phase 1 - Foundation
+- [ ] 01 - Setup
+
+## Phase 2 - Core
+- [ ] 02 - Build
+- [ ] 03 - Test
+
+## Phase 3 — Polish
+- [ ] 04 — Refine
+- [ ] 05 — Release
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				// Task 02 depends on task 01 (phase 1 → 2 boundary)
+				expect(
+					project.tasks.find((t: Task) => t.id === "02")?.dependencies,
+				).toContain("01");
+
+				// Task 04 depends on task 03 (phase 2 → 3 boundary)
+				expect(
+					project.tasks.find((t: Task) => t.id === "04")?.dependencies,
+				).toContain("03");
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("does not duplicate explicit dependencies", () => {
+			const content = `# Test
+
+## Phase 1 - MVP
+- [ ] 01 - Setup
+- [ ] 02 - Build
+
+## Phase 2 — Enhancement
+- [ ] 03 — Feature
+
+## Dependencies
+- 03 depends on 02
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				const task03 = project.tasks.find((t: Task) => t.id === "03");
+				const depCount = task03?.dependencies.filter(
+					(d: string) => d === "02",
+				).length;
+				expect(depCount).toBe(1); // Should not duplicate
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("handles single phase (no boundaries)", () => {
+			const content = `# Test
+
+## Phase 1 - All tasks
+- [ ] 01 - Task A
+- [ ] 02 - Task B
+
+## Dependencies
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				// No implicit dependencies should be added
+				expect(project.tasks[0].dependencies).toEqual([]);
+				expect(project.tasks[1].dependencies).toEqual([]);
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("works alongside explicit dependencies", () => {
+			const content = `# Test
+
+## Phase 1 - MVP
+- [ ] 01 - Setup
+- [ ] 02 - Build
+
+## Phase 2 - Enhancement
+- [ ] 03 - Feature A
+- [ ] 04 - Feature B
+
+## Dependencies
+- 04 depends on 03
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				// Task 03 has implicit dependency on task 02
+				expect(
+					project.tasks.find((t: Task) => t.id === "03")?.dependencies,
+				).toContain("02");
+
+				// Task 04 has explicit dependency on task 03
+				expect(
+					project.tasks.find((t: Task) => t.id === "04")?.dependencies,
+				).toContain("03");
+
+				// Task 04 should NOT have implicit dependency on task 02
+				expect(
+					project.tasks.find((t: Task) => t.id === "04")?.dependencies,
+				).not.toContain("02");
+			} finally {
+				cleanup();
+			}
+		});
+	});
+
+	describe("Mixed formats", () => {
+		test("phased format with arrow dependencies", () => {
+			const content = `# Test
+
+## Phase 1 - Setup
+- [ ] 01 - Initialize
+- [ ] 02 - Configure
+
+## Phase 2 - Build
+- [ ] 03 - Compile
+- [ ] 04 - Bundle
+
+## Dependencies
+- 01 → 02
+- 03 → 04
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.phases?.length).toBe(2);
+				expect(
+					project.tasks.find((t: Task) => t.id === "02")?.dependencies,
+				).toContain("01");
+				expect(
+					project.tasks.find((t: Task) => t.id === "04")?.dependencies,
+				).toContain("03");
+				expect(
+					project.tasks.find((t: Task) => t.id === "03")?.dependencies,
+				).toContain("02");
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("phased format with parallel groups", () => {
+			const content = `# Test
+
+## Phase 1 - MVP
+- [ ] 01 - Setup
+- [ ] 02 - Build
+
+## Phase 2 - Enhancement
+- [ ] 03 - Feature
+- [ ] 04 - Test
+
+## Dependencies
+- 01, 02 can be done in parallel
+- 03, 04 can be done in parallel
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.phases?.length).toBe(2);
+				expect(project.parallelGroups?.length).toBe(2);
+				expect(
+					project.tasks.find((t: Task) => t.id === "03")?.dependencies,
+				).toContain("02");
+			} finally {
+				cleanup();
+			}
+		});
+
+		test("phased format with exit criteria", () => {
+			const content = `# Test
+
+## Phase 1 - MVP
+- [ ] 01 - Build
+- [ ] 02 - Test
+
+## Phase 2 - Release
+- [ ] 03 - Deploy
+
+## Dependencies
+
+## Exit Criteria
+- All tests pass
+- Deployment successful
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				expect(project.phases?.length).toBe(2);
+				expect(project.exitCriteria?.length).toBe(2);
+			} finally {
+				cleanup();
+			}
+		});
+	});
+
+	describe("Real-world example", () => {
+		test("parses voice conversation PRD correctly", () => {
+			const content = `# Voice Conversation
+
+Objective: Add full voice conversation capability
+
+## Phase 1 - Push-to-Talk MVP
+- [ ] 01 - Build voice pipeline orchestrator → \`01-voice-pipeline-orchestrator.md\`
+- [ ] 02 - Build auto-playback audio module → \`02-auto-playback-audio-module.md\`
+- [ ] 03 - Wire voice mode toggle into chat UI → \`03-voice-mode-toggle-ui.md\`
+- [ ] 04 - End-to-end push-to-talk integration test → \`04-push-to-talk-integration-test.md\`
+
+## Phase 2 - Streaming & Real-Time
+- [ ] 05 - Build WebSocket voice channel → \`05-websocket-voice-channel.md\`
+- [ ] 06 - Implement streaming STT pipeline → \`06-streaming-stt-pipeline.md\`
+- [ ] 07 - Implement streaming TTS pipeline → \`07-streaming-tts-pipeline.md\`
+
+## Phase 3 - Optimization & Hardening
+- [ ] 08 - Model quantization and VRAM budget manager → \`08-model-quantization.md\`
+- [ ] 09 - Latency profiling and pipeline optimization → \`09-latency-profiling.md\`
+
+## Dependencies
+- 02 depends on 01
+- 03 depends on 01, 02
+- 04 depends on 03
+- 06 depends on 05
+- 07 depends on 05
+- 09 depends on 08
+
+## Exit Criteria
+- Users can hold multi-turn voice conversations
+- Total round-trip latency under 3s
+`;
+			const { project, cleanup } = parse(content);
+			try {
+				// Verify phases
+				expect(project.phases?.length).toBe(3);
+				expect(project.phases?.[0].title).toBe("Push-to-Talk MVP");
+				expect(project.phases?.[1].title).toBe("Streaming & Real-Time");
+				expect(project.phases?.[2].title).toBe("Optimization & Hardening");
+
+				// Verify task phases
+				expect(project.tasks[0].phase).toBe(1);
+				expect(project.tasks[4].phase).toBe(2);
+				expect(project.tasks[7].phase).toBe(3);
+
+				// Verify phase boundaries
+				// Task 05 (first in phase 2) depends on task 04 (last in phase 1)
+				expect(
+					project.tasks.find((t: Task) => t.id === "05")?.dependencies,
+				).toContain("04");
+
+				// Task 08 (first in phase 3) depends on task 07 (last in phase 2)
+				expect(
+					project.tasks.find((t: Task) => t.id === "08")?.dependencies,
+				).toContain("07");
+
+				// Verify explicit dependencies still work
+				expect(
+					project.tasks.find((t: Task) => t.id === "02")?.dependencies,
+				).toContain("01");
+				expect(
+					project.tasks.find((t: Task) => t.id === "03")?.dependencies,
+				).toContain("01");
+				expect(
+					project.tasks.find((t: Task) => t.id === "03")?.dependencies,
+				).toContain("02");
+
+				// Verify task files
+				expect(project.tasks[0].file).toBe("01-voice-pipeline-orchestrator.md");
+				expect(project.tasks[1].file).toBe("02-auto-playback-audio-module.md");
+
+				// Verify exit criteria
+				expect(project.exitCriteria?.length).toBe(2);
+				expect(project.objective).toBe("Voice Conversation");
+			} finally {
+				cleanup();
+			}
+		});
+	});
+});

tsconfig.json

@@ -1,18 +1,15 @@
 {
   "compilerOptions": {
     "target": "ES2022",
-    "module": "commonjs",
+    "module": "ES2022",
+    "moduleResolution": "bundler",
     "lib": ["ES2022"],
-    "outDir": "./dist",
-    "rootDir": "./",
+    "noEmit": true,
     "strict": true,
     "esModuleInterop": true,
     "skipLibCheck": true,
     "forceConsistentCasingInFileNames": true,
-    "resolveJsonModule": true,
-    "declaration": true,
-    "declarationMap": true,
-    "sourceMap": true
+    "resolveJsonModule": true
   },
   "include": ["index.ts", "src/**/*"],
   "exclude": ["node_modules", "dist"]