From 008489a91b0dd52039a4df76d1e1ba329232ab8e Mon Sep 17 00:00:00 2001
From: Michael Freno <michaelt.freno@gmail.com>
Date: Fri, 17 Jul 2026 11:11:21 -0400
Subject: [PATCH] fix: pass parent modelRuntime so extension providers work in
 ralpi sessions

ralpi's runAgentSession was letting createAgentSession create a fresh
ModelRuntime from models.json only. This excluded extension-registered
providers like neuralwatt, whose streamSimple wrapper handles:
- 429 rate-limit header parsing (concurrent/TPM/admission/RPM)
- Rate-limit error normalization for Pi's retry logic
- Context-overflow error normalization

Without the wrapper, neuralwatt rate-limit and context-overflow errors
surfaced as raw unrecoverable errors instead of being retried by Pi's
built-in retry. Now passes ctx.modelRegistry.runtime (the underlying
ModelRuntime) through to createAgentSession.
---
 src/executor.ts | 9 ++++++++-
 src/utils.ts    | 7 +++++++
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/src/executor.ts b/src/executor.ts
index 654ac79..419e00d 100644
--- a/src/executor.ts
+++ b/src/executor.ts
@@ -3,7 +3,10 @@ import * as path from "node:path";
 import type { Task, Project, Reflection, ToolUsage } from "./types";
 import type { RalpiConfig } from "./types";
 import type { ProgressTracker } from "./progress";
-import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
+import type {
+	ExtensionContext,
+	ModelRuntime,
+} from "@earendil-works/pi-coding-agent";
 import { buildTaskPrompt, buildReviewPrompt, MAX_DIFF_BYTES } from "./prompts";
 import { extractReflection } from "./reflection";
 import {
@@ -291,6 +294,8 @@ export async function runTask(
 		undefined, // no abort signal
 		assignedModel ?? config.model,
 		config.thinkingLevel,
+		false, // noSkills — task sessions need skills
+		(ctx.modelRegistry as any).runtime as ModelRuntime,
 	);
 
 	const durationMs = Date.now() - startMs;
@@ -713,6 +718,7 @@ async function executeTask(
 								"",
 								"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.",
+								"Do NOT include the task number, task ID, or any ralpi task reference in the commit message. The commit message must describe only the work done — never mention the task ID (e.g. `task 03`, `#3`, etc.).",
 								"",
 								"### Current Changes (git status --porcelain)",
 								"```text",
@@ -1061,6 +1067,7 @@ async function runFollowUpSession(
 				model,
 				config.thinkingLevel,
 				true, // noSkills — follow-up sessions don't need the skills catalog
+				(ctx.modelRegistry as any).runtime as ModelRuntime,
 			);
 
 			if (result.success) break;
diff --git a/src/utils.ts b/src/utils.ts
index 1b6ac23..00c8afb 100644
--- a/src/utils.ts
+++ b/src/utils.ts
@@ -14,6 +14,7 @@ import {
 	getAgentDir,
 	SessionManager,
 	SettingsManager,
+	type ModelRuntime,
 } from "@earendil-works/pi-coding-agent";
 
 // ─── Directory Helpers ───────────────────────────────────────────────────────
@@ -512,6 +513,11 @@ export async function runAgentSession(
 	 *  focused follow-up sessions (commit/review) that don't need skills —
 	 *  keeps the context lean and avoids dragging in unrelated overhead. */
 	noSkills = false,
+	/** Parent session's model runtime. Must be passed so extension-registered
+	 *  providers (e.g., neuralwatt with its streamSimple wrapper for 429
+	 *  rate-limit normalization) are available. When omitted, the SDK creates
+	 *  a fresh runtime from models.json only — extension providers are lost. */
+	modelRuntime?: ModelRuntime,
 ): Promise<{
 	success: boolean;
 	text: string;
@@ -556,6 +562,7 @@ export async function runAgentSession(
 			sessionManager: SessionManager.inMemory(),
 			resourceLoader: loader,
 			settingsManager: SettingsManager.create(cwd, getAgentDir()),
+			modelRuntime,
 			tools: ["read", "bash", "edit", "write", "grep", "find", "ls"],
 			model: model as any,
 			thinkingLevel: thinkingLevel as any,