1 Commits

Author SHA1 Message Date
2fd2a4a71a security: stop trusting the leftmost x-forwarded-for for rate limits and abuse quotas (FRE-371)
getClientIP (the rate-limit and audit key) took the LEFTMOST x-forwarded-for
hop and read x-real-ip unconditionally. In our topology — client → Cloudflare
→ same-host origin proxy → app — both proxies APPEND, so the leftmost hop is
whatever the sender typed: a rotating X-Forwarded-For minted a fresh rate-limit
bucket per request, and unlimited bug-report quota via the same pattern in
routes/api/the-nook/report.ts. x-real-ip is client-settable too; nothing in
the chain overwrites it. It is no longer read at all.

New trust order (src/server/trusted-client-ip.ts, resolveTrustedClientIp):

  1. a non-loopback socket peer IS the client — its headers are ignored;
  2. cf-connecting-ip (Cloudflare overwrites it at the edge; parsed as an
     address or rejected);
  3. x-forwarded-for counted from the RIGHT, discounting one hop for the
     Cloudflare edge address the origin proxy appended (the topology has
     exactly one trusted proxy below Cloudflare);
  4. "unknown" — callers key stricter shared limits on it, so resolution
     fails closed toward a tighter bucket, never a looser one.

Addresses are never truncated: a /24-truncated key would pool attackers with
the innocents behind them.

Share-vs-copy, stated deliberately: this is a port of the parsing primitives
 FRE-366 reviewed in request-client.ts, NOT an import. resolveClientIp carries
a standing rule (1d9287a) that it never feeds a security decision; lifting
that rule for these callers would couple an access-control change to an
analytics branch under review, and the analytics helper only exposes
truncated prefixes while rate limiting needs the full address. If the two
drift, trusted-client-ip.ts is the one security paths follow. The deploy-side
precondition (origin reachable only via Cloudflare + the same-host proxy) is
documented in the file header; if the app is ever directly internet-reachable
no in-app read can recover the true client.

Tests updated deliberately, not to go green: rate-limit.test.ts now pins the
rightmost-minus-one order, cf-connecting-ip precedence, peer-over-headers,
x-real-ip rejection, and the spoof case (a rotating leftmost hop does not
change the resolved client); report.test.ts pins the quota key stability
under a rotating header. New trusted-client-ip.test.ts covers the helper's
own edges: APIEvent wrapper, loopback forms, unparseable hops failing closed.
2026-10-30 05:06:54 +00:00
15 changed files with 635 additions and 689 deletions

View File

@@ -4,39 +4,13 @@ Every drafted asset draws its voice, claims, and constraints from this file. If
artifact invents its own framing, fix the artifact or fix this file — never both
silently diverging.
**Reconciled 2026-10-29**, and again **2026-11-03** for the FRE-997 correction,
against the CTO's verdicts ([FRE-315], [FRE-322], [FRE-354], [FRE-997]), the Chief
of Staff's price decision (2026-10-29 04:01), the Plan B launch route ([FRE-346] →
[FRE-203] `plan-b`), and `The-Nook` master (app floor is **14.2**, not 14.0 —
`freno-dev` `881190d`, site merge `134f41e`). The previous version of this file
named an upstream `the-nook/docs/marketing-strategy.md` that no longer exists; the
verdicts below are the source of truth it was pointing at.
[FRE-315]: https://paperclip.freno.me/FRE/issues/FRE-315
[FRE-322]: https://paperclip.freno.me/FRE/issues/FRE-322
[FRE-354]: https://paperclip.freno.me/FRE/issues/FRE-354
[FRE-320]: https://paperclip.freno.me/FRE/issues/FRE-320
[FRE-346]: https://paperclip.freno.me/FRE/issues/FRE-346
[FRE-203]: https://paperclip.freno.me/FRE/issues/FRE-203
[FRE-997]: https://paperclip.freno.me/FRE/issues/FRE-997
Primary source: `~/Code/the-nook/docs/marketing-strategy.md`.
## The sentence
There is no single approved public sentence yet, and saying otherwise is the one
mistake this file exists to prevent. What exists is a **launch claim set** — two
sentences, both from one 15-second take, recorded as the chosen route by the Chief
of Staff on 2026-10-29 ([FRE-346] `decision` §2–§4):
| Slot | Exact sentence | Status today |
|---|---|---|
| Launch line | **"Every agent you're running reports into one notch, and the one waiting on you pins above the rest."** | Verified by code ([FRE-322] §3). Needs the one-Mac take before it is publishable. |
| Second line, same take | **"Approve an agent's tool call from your Mac's notch, and the agent carries on."** | Needs the demo ([FRE-322] §1). The same take settles it. |
Working line for internal writing, not for publication:
> **Your Mac and your agents, under control.** One notch for the whole fleet: you
> see what is running, you unblock what is waiting, and the machine stays cool
> while it works.
> **Your Mac and your agents, under control.** The Nook puts your whole coding-agent
> fleet in the notch — thirteen agents, local or over SSH — so you see what is running,
> unblock it in one tap, and keep the machine from overheating while it works.
## The wedge
@@ -48,18 +22,15 @@ The thing to say over and over, in every channel, in slightly different words:
It is acute, quantifiable, and true — and it is the demo: a card appears, you tap
Allow, work resumes. Ten seconds, no explanation needed.
## The consolidation reframe
## The bundling reframe
Three audiences (agent herders, hardware tinkerers, glanceable-widget users) in one
app reads like a grab bag. Reframe as consolidation, which is the honest story:
$10 app reads like a grab bag. Reframe as consolidation, which is the honest story:
> One app instead of a notch toy, a fan utility, a battery limiter, a menu-bar stats
> strip, and a notification hack. One purchase, no subscription.
> One $10 app instead of a notch toy, a fan utility, a battery limiter, a menu-bar
> stats strip, and a notification hack. Same one-time purchase, no subscription.
**No price, no discount and no future price in copy.** The price is $10, it stays
$10, and the dated rise to $15 is withdrawn (Chief of Staff, 2026-10-29 04:01). A
price change is a Chief of Staff + sponsor call; it is never a marketing decision
and never a launch asset.
Price rises to $15 at 1.0 — dated and honest, not a permanent beta.
## Hard rules
@@ -77,54 +48,24 @@ pages, marketing copy, comparison tables, meta descriptions, emails, FAQs, all o
community replies written by a human, where quoting what a user said about another
tool is normal conversation.
### 2. A claim needs a verdict, not just a test
### 2. A claim requires a test to exist
Copy that says "every agent", "13 agents", "no data leaves the device" must point at
both a test **and** a CTO verdict naming the exact sentence. A passing test is not a
verdict; a verdict is not a recording. If a sentence cannot be traced to a row in the
tables below, it does not go in copy.
Copy that says "13 agents" or "no data leaves the device" must have a test backing it.
If a claim cannot be pointed at code or a test, it does not go in copy.
### 3. Honest privacy scoping
Telemetry and the licence check **do** leave the device: Sentry runs on **every
launch, unconditionally** (with the Mac's IP address via `sendDefaultPii`), and the
trial check sends `IOPlatformUUID` and the device name to `freno.me` ([FRE-322]
§2.4, corrected by [FRE-354] `(b)`). Never publish, in any wording:
Telemetry and the trial email **do** leave the device. The defensible claim is:
- "no data leaves the device" / "local-only" / "no telemetry" / "no account"
- "nothing leaves your Mac"
> **No prompt or code content leaves the device.**
The one favourable engine sentence — about the **hook and bridge only**, never
"The Nook" — as corrected by [FRE-354] `(j)`: the hook and bridge **are not a cloud
service**. Do not write "talk to nothing but your Mac" (too absolute: the
`NOOK_PORTS` fan-out resolves `host.docker.internal` and the default gateway, and
`defaultGateway()` opens a no-packet UDP socket to `8.8.8.8:80`), and do not narrow
it to "over SSH" (a tailnet peer dials directly).
It is verified as a code fact ([FRE-322] §2, [FRE-354] `(j)`), and it is still
**held** until the disclosures land: Sentry gated behind a preference or removed,
and `/privacy` + `/faq` corrected against the egress inventory (patch on branch
`faq/from-objections-2026-10-29`, verified by [FRE-354]). A true sentence next to
false ones is worse than no sentence.
Never "no data leaves the device".
### 4. Sell what works today
An iPhone + Apple Watch companion **is** in users' hands as a TestFlight beta
(`ios-v0.3.2`, 2026-10-22; [FRE-354] open question). That makes one thing true and
one thing forbidden: say "TestFlight beta" when the phone exists in a sentence at
all, and never claim approvals-from-phone as a shipped capability without its own
CTO verdict — no "(coming soon)" parentheticals either way.
Remote SSH is **publishable**, and the site sells it again ([FRE-997]). The
copy that was pulled on 2026-10-29 had been checked against claims it never made:
hosts come from `~/.ssh/config` and Tailscale (`RemoteHostDiscovery`), the
one-time `ssh-copy-id` is a prerequisite the row hands the user as a command
(`RemoteHostStatus.fixCommand`), the per-host credential is minted and copied
over (`RemoteMirror`), and a remote host mirrors this Mac's integrations
(`RemoteMirrorReconcile`). Two things in the old copy were wrong and were fixed
rather than restored: the "every hooked agent" overstatement, and a count that
said thirteen against a registry of eighteen. Muse, Cline and Amp are
`.unsupported` for remote in `AgentSpec.remote`, and the copy names them.
Phone approvals are deferred until a shipped iOS app exists — no "(coming soon)"
parentheticals. The launch story is remote SSH: agents running on other machines,
visible and unblockable from the Mac's notch.
### 5. Humans hold every account
@@ -132,30 +73,23 @@ No agent authenticates, posts, upvotes, or DMs. No sock puppets, no self-mention
without disclosure. `approved/` **is the only thing a human ever publishes from** —
nothing posts itself.
## Claims: status at 2026-11-03
## Approved claims (each backed by code)
Each row names the evidence, not a feeling. Re-read the verdict before using a row.
| Claim | Status | Evidence / next step |
|---|---|---|
| Approve an agent's tool call from the notch, and it carries on | **Needs demo** — not publishable | [FRE-322] §1. `AgentLiveTurnTests` has never run on a signed-in host; one observed local run settles it. Say "an agent", never "any agent", never "always". |
| Every agent you're running reports into one notch; the blocked one pins above the rest | **Verified by code, gated on the take** | [FRE-322] §3. Same one-Mac take as above. No count in the sentence. |
| The hook and bridge are not a cloud service | **Verified code fact, held on disclosure** | [FRE-322] §2, wording corrected by [FRE-354] `(j)`. Held until Sentry is gated/removed and `/privacy` + `/faq` land. Never widen to "The Nook", never "nothing but your Mac". |
| Fan, thermal and charge control | **No claim verdict yet — do not lead with it** | Backed by `helper/` (privileged helper, SMC, no kext), but no CTO sentence exists for it. Documentary, not launch. |
| One-time purchase, no subscription | **Publishable fact, price not a launch asset** | $10, held at $10. No dated rise, no discount framing, no pricing copy. |
| Agent coverage numbers ("eighteen agents", "18 agents") | **Publishable at 18** | [FRE-997]: the /agents table now lists every row of `AgentSpec.all` and `agents.test.ts` asserts equality with the 18-name registry in registry order, so the landing bullet, the marquee chip and the /agents copy say 18. No other number is publishable, and `every` / `all` are still banned without naming the three local-only agents. |
| macOS 14.2 or later | **Publishable fact** | The app cannot start below 14.2 (`freno-dev` `881190d`, site `134f41e`). "macOS 14+" is false and was fixed on the site. |
| Remote agents over SSH, one click to install | **Publishable, with the three local-only agents named** | [FRE-997]: `RemoteHostDiscovery` (ssh config + tailnet), `RemoteMirror` (hook push, credential, mirroring), `RemoteMirrorReconcile`, `RemoteHostStatus.fixCommand` (the one-time `ssh-copy-id` the row shows). `AgentSpec.remote` is `.unsupported` for Muse, Cline and Amp — the landing bullet and /agents paragraph name them; never say "every hooked agent". |
| "Nothing leaves your Mac", "no telemetry", "local-only", "no account" | **False today — banned** | Sentry unconditional, licence check with hardware fingerprint ([FRE-322] §2.4). |
| Approvals from the iPhone or Apple Watch | **No verdict — not a launch claim** | The companion is a TestFlight beta ([FRE-354]); the approval-from-phone claim has no CTO verdict and stays out of launch copy. |
| Claim | Backing |
|---|---|
| Thirteen hooked agents, one glance | `AgentSpec.all` — claude, codex, opencode, pi, omp, muse, hermes, openclaw, zai, gemini, cline, qwen, deepseek |
| Remote agents over SSH | `hook/setup.go` remote install path; `NOOK_HOST` TCP transport |
| Permission gate answered from a card | awaitPolicyTable + bridge directive path (`hook/normalize.go`, `serialize.go`) |
| Fan / thermal / charge control | `helper/` (privileged helper, SMC, no kext) |
| One-time purchase, no subscription | $10 → $15 at 1.0; Stripe checkout |
| macOS 14+ | 27 source files use the `@Observable` framework |
| No prompt or code content leaves the device | envelope carries session/event/terminal metadata only (see `docs/agent-interface.md`) |
## Voice
- Short, concrete, no hype adjectives. Numbers over adjectives — the only count
that ships is eighteen, and it is the registry's own size (see the coverage row
above).
- The notch is the product's home; agents are the product's point.
- Short, concrete, no hype adjectives. Numbers over adjectives ("thirteen agents",
"one tap", "$10 once").
- The notch/island is the product's home; agents are the product's point.
- Never apologize for being a utility. The utility is the point.
- Banned phrasings: competitor names (see rule 1), "AI-powered", "revolutionary",
"game-changer", "seamlessly", any unverifiable superlative — and the specific
banned sentences in rule 3 and the table above.
"game-changer", "seamlessly". Unverifiable superlatives generally.

View File

@@ -959,7 +959,7 @@ export default function FeatureBreakdowns() {
title="Your whole fleet, in one pixel strip"
body="Every session, every agent, one square each. The pill never wraps, never grows — an arriving question flashes the fire and swaps the slot, then hands the notch back."
bullets={[
"18 agents: Claude Code, Codex, Gemini CLI, OpenCode, Pi, Muse, and more",
"13 agents: Claude Code, Codex, Gemini CLI, OpenCode, Pi, Muse, and more",
"Custom agents: your own coding agent can wire itself in from the published contract",
"Slot rankings — you choose which data earns the notch",
"You don't need a notch — external displays get the same island, floating top-center"
@@ -976,7 +976,7 @@ export default function FeatureBreakdowns() {
"Approval cards pin above the session list; questions answer inside their row",
"Allow for session waves through the rest of that session's requests",
"Answer from iPhone or Apple Watch too, now in TestFlight beta",
"Works across local and remote sessions; Muse, Cline and Amp are this Mac only"
"Works across every hooked agent, local or remote"
]}
reversed
>

View File

@@ -722,7 +722,7 @@ function LayoutMock() {
/* ── Marquee ─────────────────────────────────────────────────────── */
const CHIPS_A = [
"18 agents",
"13 agents",
"Custom agents",
"Six chiptunes",
"Focus-aware slots",

View File

@@ -10,6 +10,8 @@ import { describe, expect, it, mock, beforeEach } from "bun:test";
const sentEmails: { to: string; subject: string; html: string }[] = [];
const tokenAnswers: { allowed: boolean; retryAfterSec?: number }[] = [];
/** Every rate-limit key the route derived, in order. */
const rateKeys: string[] = [];
mock.module("~/server/email", () => ({
default: async (to: string, subject: string, html: string) => {
@@ -19,7 +21,10 @@ mock.module("~/server/email", () => ({
}));
mock.module("~/server/bug-report-rate-limit", () => ({
takeBugReportToken: () => tokenAnswers.shift() ?? { allowed: true }
takeBugReportToken: (key: string) => {
rateKeys.push(key);
return tokenAnswers.shift() ?? { allowed: true };
}
}));
const { POST } = await import("./report");
@@ -72,6 +77,7 @@ describe("POST /api/the-nook/report", () => {
beforeEach(() => {
sentEmails.length = 0;
tokenAnswers.length = 0;
rateKeys.length = 0;
});
it("emails the owner a report containing the snapshot, escaped", async () => {
@@ -133,10 +139,49 @@ describe("POST /api/the-nook/report", () => {
it("still succeeds when no fingerprint header exists", async () => {
const res = await POST(
request(validPayload, { "x-forwarded-for": "203.0.113.9, 10.0.0.1" })
request(validPayload, {
"cf-connecting-ip": "203.0.113.9",
"x-forwarded-for": "203.0.113.9, 10.0.0.1"
})
);
expect(res.status).toBe(200);
expect(sentEmails).toHaveLength(1);
// The quota key is the Cloudflare-recorded client, not a header hop the
// sender chose.
expect(rateKeys).toEqual(["203.0.113.9"]);
});
it("a rotating leftmost x-forwarded-for does not mint a fresh quota key", async () => {
// FRE-371: the old read keyed on the leftmost hop, which the sender
// controls outright — one scripted loop with a rotating header had
// unlimited reports. With the same Cloudflare-recorded client and a
// different spoofed prefix per request, the key must be stable.
for (const spoof of ["1.1.1.1", "2.2.2.2", "3.3.3.3"]) {
const res = await POST(
request(validPayload, {
"cf-connecting-ip": "203.0.113.9",
"x-forwarded-for": `${spoof}, 203.0.113.9, 10.0.0.1`
})
);
expect(res.status).toBe(200);
}
expect(rateKeys).toEqual([
"203.0.113.9",
"203.0.113.9",
"203.0.113.9"
]);
});
it("keys on the rightmost-minus-one hop when Cloudflare's header is absent", async () => {
const res = await POST(
request(validPayload, {
"x-forwarded-for": "6.6.6.6, 203.0.113.9, 10.0.0.1"
})
);
expect(res.status).toBe(200);
// Never the leftmost (sender-chosen) hop.
expect(rateKeys).toEqual(["203.0.113.9"]);
});
it("delivers the landing page's beta request with only the email", async () => {

View File

@@ -3,6 +3,7 @@ import { z } from "zod";
import sendEmail from "~/server/email";
import { CONTACT_RECIPIENT_EMAIL } from "~/lib/contact-config";
import { takeBugReportToken } from "~/server/bug-report-rate-limit";
import { resolveTrustedClientIp } from "~/server/trusted-client-ip";
import { json, error } from "./_lib";
/**
@@ -103,10 +104,9 @@ export async function POST(event: APIEvent): Promise<Response> {
)
.trim()
.toLowerCase();
const forwarded = event.request.headers.get("x-forwarded-for");
const rateKey =
fingerprint ||
(forwarded ? forwarded.split(",")[0]!.trim() : "unknown");
// FRE-371: the IP fallback must be the spoof-resistant read — a rotating
// leftmost x-forwarded-for used to mint unlimited report quota.
const rateKey = fingerprint || resolveTrustedClientIp(event);
const token = takeBugReportToken(rateKey);
if (!token.allowed) {
return new Response(JSON.stringify({ error: "Too many reports" }), {

View File

@@ -1,5 +1,5 @@
/**
* Regression test for the /agents roster and its coverage count.
* Regression test for the /agents roster (NOOK-336).
*
* The page mirrors the app's registry, `AgentSpec.all` in
* the-nook/Sources/NookCore/Agents/AgentRegistry.swift. When Muse joined
@@ -8,12 +8,6 @@
* them. The registry lives in another repo, so the expected roster is
* restated here; when an agent is added there, add it here and on the page.
*
* FRE-997 restored the count. The page said thirteen while the registry held
* eighteen, so the count was dropped rather than corrected (FRE-359 verdict B);
* now that all eighteen rows are on the page, the roster check is an equality
* check again and the page states eighteen. A subset check would let a shipped
* agent go missing behind a passing suite, which is the NOOK-336 failure.
*
* STATIC SOURCE AUDIT, the same pattern as the downloads regressions.
*/
import { describe, it, expect } from "bun:test";
@@ -22,13 +16,7 @@ import { join } from "node:path";
const SOURCE = readFileSync(join(import.meta.dir, "agents.tsx"), "utf8");
/** User-facing copy only: comments are not copy and may name the old wording. */
const COPY = SOURCE.replace(/\/\*[\s\S]*?\*\//g, "").replace(/^\s*\/\/.*$/gm, "");
/** JSX wraps prose across lines; match a sentence, not its indentation. */
const PROSE = COPY.replace(/\s+/g, " ");
/** AgentSpec.all display names, in registry order (18 rows). @ The-Nook f4258034, 2026-10-29 */
/** AgentSpec.all display names, in registry order. */
const REGISTRY = [
"Claude Code",
"Codex",
@@ -42,49 +30,25 @@ const REGISTRY = [
"Gemini CLI",
"Cline",
"Qwen Code",
"DeepSeek",
"Kimi",
"GitHub Copilot CLI",
"Crush",
"Cursor",
"Amp"
"DeepSeek"
];
const NUMBER_WORDS = ["twelve", "thirteen", "fourteen", "fifteen"];
const entries = [...SOURCE.matchAll(/^\s{4}name: "([^"]+)",$/gm)].map((m) => m[1]);
describe("Nook /agents roster", () => {
it("lists every registry agent, once, in registry order", () => {
describe("Nook /agents roster (NOOK-336)", () => {
it("lists every registry agent, in registry order", () => {
expect(entries).toEqual(REGISTRY);
});
it("states the registry's size", () => {
expect(entries.length).toBe(18);
expect(COPY).toContain("all eighteen");
expect(COPY).toContain("The 18 agents");
expect(COPY).toContain("eighteen coding agents");
});
it("carries no stale count", () => {
expect(COPY.toLowerCase()).not.toContain("thirteen");
expect(COPY).not.toMatch(/\b13 agents\b/);
expect(COPY).not.toContain("all thirteen");
});
});
describe("Nook /agents remote coverage", () => {
it("says agents can run on remote machines over SSH", () => {
expect(PROSE).toContain("remote machines");
expect(PROSE).toContain("remote sessions report into the same grid");
});
it("names the three agents that stay on this Mac", () => {
expect(PROSE).toContain("Muse, Cline and Amp run on this Mac only");
});
it("discloses the local-only rows where they are listed", () => {
const localOnly = [
...SOURCE.matchAll(/This Mac only — remote hosts skip it/g)
];
expect(localOnly.length).toBe(3);
it("states the same count everywhere it names one", () => {
const word = NUMBER_WORDS[entries.length - 12];
for (const other of NUMBER_WORDS.filter((w) => w !== word)) {
expect(SOURCE.toLowerCase()).not.toContain(other);
}
expect(SOURCE).toContain(`all ${word}`);
expect(SOURCE).toContain(`The ${entries.length} agents`);
expect(SOURCE).toContain(`watches ${word} coding agents`);
});
});

View File

@@ -1,10 +1,9 @@
/**
* `/agents` on the Nook subdomain — all eighteen agents in the registry and
* each row's event coverage.
* `/agents` on the Nook subdomain — supported agents and their event coverage.
*
* The registry below mirrors `AgentSpec.all` in the app
* (~/Code/the-nook/Sources/NookCore/Agents/AgentRegistry.swift) and each
* agent's installed hook surface (hook/setup.go) or plugin event subscription
* (~/Code/the-nook/Sources/NookCore/Agents/AgentSpec.swift) and each agent's
* installed hook surface (hook/setup.go) or plugin event subscription
* (plugins/*), normalized onto the shared event vocabulary
* (~/Code/the-nook/docs/agent-interface.md). Copy follows
* marketing/positioning.md: category nouns, honest scoping, no unverifiable
@@ -209,8 +208,7 @@ const AGENTS: AgentEntry[] = [
},
{
name: "Cline",
wiring:
"One hook file per event in its hooks directory. This Mac only — remote hosts skip it",
wiring: "One hook file per event in its hooks directory",
events: [
"SessionStart",
"SessionEnd",
@@ -261,93 +259,6 @@ const AGENTS: AgentEntry[] = [
],
gating:
"Tool calls can be allowed or denied from a card; permission prompts stay with the agent"
},
{
name: "Kimi",
wiring: "Hook rules in its TOML config, one per event",
events: [
"SessionStart",
"SessionEnd",
"UserPromptSubmit",
"AgentStart",
"PreToolUse",
"PermissionRequest",
"PermissionResolved",
"PostToolUse",
"PostToolUseFailure",
"Stop",
"SessionError",
"SubagentStart",
"SubagentStop",
"PreCompact",
"Notification"
],
gating:
"Observation — every event appears as a card; its PreToolUse fires before its own permission check, so the agent keeps its own prompting"
},
{
name: "GitHub Copilot CLI",
wiring: "One hook file of ours in its hooks directory",
events: [
"SessionStart",
"SessionEnd",
"UserPromptSubmit",
"PreToolUse",
"PermissionRequest",
"PostToolUse",
"PostToolUseFailure",
"Stop",
"SubagentStart",
"SubagentStop",
"ErrorOccurred",
"PreCompact",
"Notification"
],
gating:
"Observation — every event appears as a card; its permission request runs before its own policy check, so the agent keeps its own prompting"
},
{
name: "Crush",
wiring:
"One entry in its JSONC config, the only event it documents",
events: ["PreToolUse"],
gating:
"Observation — tool calls appear as cards; its own permission prompts decide"
},
{
name: "Cursor",
wiring:
"Hook entries in its hooks.json, merged with your own. No permission hook, on purpose",
events: [
"SessionStart",
"SessionEnd",
"UserPromptSubmit",
"PostToolUse",
"PostToolUseFailure",
"Stop",
"SessionError",
"AfterAgentResponse",
"SubagentStop",
"PreCompact"
],
gating:
"Observation — a permission hook that stays silent blocks the action here, so Cursor keeps its own prompting"
},
{
name: "Amp",
wiring:
"In-process extension package. This Mac only — remote hosts skip it",
events: [
"SessionStart",
"SessionEnd",
"UserPromptSubmit",
"PreToolUse",
"PostToolUse",
"Stop",
"SessionError"
],
gating:
"Observation — its plugin API has no permission-wait event, so the agent keeps its own prompting"
}
];
@@ -372,8 +283,8 @@ export default function NookAgentsPage() {
<EdgeCacheHeaders maxAge={300} />
<PageHead
title="Supported agents"
description="Every coding agent The Nook supports — all eighteen, with each hook's exact event coverage and what you can do from a card: watch sessions live, answer permission prompts, unblock work."
ogTitle="The 18 agents The Nook supports, and what each hook sees"
description="Every coding agent The Nook supports — all thirteen, with each hook's exact event coverage and what you can do from a card: watch sessions live, answer permission prompts, unblock work."
ogTitle="The 13 agents The Nook supports, and what each hook sees"
/>
<SubdomainHeader />
<main class="min-h-screen px-[8vw] py-[8vh]">
@@ -381,7 +292,7 @@ export default function NookAgentsPage() {
Supported agents
</h1>
<p class="text-subtext0 mb-2 max-w-2xl text-lg">
The Nook watches eighteen coding agents in one place — your menu bar and
The Nook watches thirteen coding agents in one place — your menu bar and
notch show every session that is running, thinking, or blocked, and a
card answers permission prompts without leaving your keyboard.
</p>
@@ -390,8 +301,7 @@ export default function NookAgentsPage() {
means the agent hands the decision to The Nook while it waits;
"Observe" means every event still appears live, but the agent keeps
its own prompting. Agents can run on this Mac or on remote machines
over SSH — remote sessions report into the same grid. Muse, Cline and
Amp run on this Mac only.
over SSH — remote sessions report into the same grid.
</p>
<div class="grid gap-4 lg:grid-cols-2">

View File

@@ -1,112 +0,0 @@
/**
* Guard for the remote / SSH story on the Nook site, and for the two phrasings
* that must not come back with it (FRE-323 → FRE-997).
*
* FRE-323 removed the remote copy because it had been checked against claims
* the copy never made: the app does list hosts from `~/.ssh/config` and
* Tailscale, a one-time `ssh-copy-id` is a normal prerequisite the row explains,
* and the per-host credential is minted and copied automatically
* (`RemoteHostDiscovery`, `RemoteHostStatus.fixCommand`, `RemoteMirror`). Two
* things in the old copy were genuinely wrong, and this file guards those
* instead of the copy as a whole:
*
* - "Works across every hooked agent, local or remote" overstated it. Muse,
* Cline and Amp are `.unsupported` for remote in `AgentSpec.remote`, so the
* landing bullet and the /agents paragraph name them.
* - The count said thirteen while `AgentSpec.all` holds eighteen. The fix was
* to say eighteen, not to drop the count.
*
* So the removal guards from `d44133f` are gone: a removed sentence is exactly
* what this task restores. What stays is the pair of bans, because either one
* coming back would repeat the original defect.
*
* STATIC SOURCE AUDIT, the same pattern as `agents.test.ts` and
* `downloads.test.ts` (SolidJS does not render under `bun:test`).
*/
import { describe, it, expect } from "bun:test";
import { readFileSync } from "node:fs";
import { join } from "node:path";
const ROOT = join(import.meta.dir, "../../..");
/** Shipped page and component source. */
const AUDITED = [
"src/routes/nook/index.tsx",
"src/routes/nook/agents.tsx",
"src/routes/nook/faq.tsx",
"src/components/nook/FeatureCollage.tsx",
"src/components/nook/FeatureBreakdowns.tsx",
"src/components/nook/IslandExtras.tsx"
];
/** The overstatements the restored copy must never carry again. */
const BANNED_REMOTE_CLAIMS = [
// a universal claim: three registry agents cannot mirror over SSH
"every hooked agent, local or remote",
"Works across every hooked agent",
// the stale count this task replaced
"13 agents",
"all thirteen",
"thirteen coding agents"
];
/** User-facing copy only: comments are not copy and may name the old wording. */
function copyOf(file: string): string {
return readFileSync(join(ROOT, file), "utf8")
.replace(/\/\*[\s\S]*?\*\//g, "")
.replace(/^\s*\/\/.*$/gm, "");
}
describe("the restored remote copy is on the site (FRE-997)", () => {
it("the landing collage sells agents on other machines", () => {
const copy = copyOf("src/components/nook/FeatureCollage.tsx");
for (const phrase of [
"Agents on other machines",
"one click installs the nook-hook",
"Remote"
]) {
expect(copy).toContain(phrase);
}
});
it("the landing breakdown keeps the SSH section and its demo", () => {
const copy = copyOf("src/components/nook/FeatureBreakdowns.tsx");
for (const phrase of [
"Any SSH box, one click away",
"Pick a host from your ssh config or your tailnet and click once",
"Remote sessions are ordinary rows",
"RemoteHostDemo"
]) {
expect(copy).toContain(phrase);
}
});
it("the chip is back", () => {
expect(copyOf("src/components/nook/IslandExtras.tsx")).toContain(
"One-click SSH setup"
);
});
it("/agents says remote sessions report into the same grid", () => {
const copy = copyOf("src/routes/nook/agents.tsx");
expect(copy).toContain("remote machines");
expect(copy).toContain("remote sessions report into the same grid");
});
});
describe("the two original overstatements stay off the site (FRE-323, FRE-359)", () => {
for (const file of AUDITED) {
it(`${file} claims no universal remote coverage and no stale count`, () => {
const copy = copyOf(file);
for (const phrase of BANNED_REMOTE_CLAIMS) {
expect(copy.toLowerCase()).not.toContain(phrase.toLowerCase());
}
});
}
it("the landing names the agents that stay local", () => {
expect(copyOf("src/components/nook/FeatureBreakdowns.tsx")).toContain(
"Muse, Cline and Amp are this Mac only"
);
});
});

View File

@@ -1,139 +0,0 @@
/**
* A Smarter Remote privacy policy — `freno.me/smarter-remote/privacy`.
*
* Linked from the iOS app's paywall (App Review 3.1.2). Reflects what the app
* actually does: TV control stays on the local network, pairing data stays on
* the phone, purchases go through Apple and RevenueCat, and sharing is a QR
* code passed between phones.
*/
import { A } from "@solidjs/router";
import { PageHead } from "~/components/PageHead";
import { EdgeCacheHeaders } from "~/components/EdgeCacheHeaders";
export default function SmarterRemotePrivacyPolicy() {
return (
<>
<EdgeCacheHeaders maxAge={300} />
<PageHead
title="A Smarter Remote Privacy Policy"
description="Privacy policy for A Smarter Remote, a universal TV remote app that works over your local Wi-Fi network."
/>
<div class="bg-base">
<div class="min-h-screen px-[8vw] py-[8vh]">
<div class="py-4 text-xl">A Smarter Remote&apos;s Privacy Policy</div>
<div class="py-2">Last Updated: November 2, 2026</div>
<div class="py-2">
Welcome to A Smarter Remote (&apos;We&apos;, &apos;Us&apos;,
&apos;Our&apos;). The app controls your TVs over your own Wi-Fi
network. It has no accounts, no ads, and no analytics. This policy
explains the little data it does handle.
</div>
<ol>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">1.</span> Data We Collect
</div>
<div class="pl-4">
<div class="pb-2">
<div class="-ml-6">(a) Your TVs:</div> The app finds TVs on
your local network and stores their names, network addresses,
and platform on your phone. Pairing codes and tokens are kept
in your phone&apos;s Keychain. This data is sent only to the
TV itself, over your local network. It never reaches us.
</div>
<div class="pb-2">
<div class="-ml-6">(b) Purchases:</div> Payments are handled
by Apple. We use RevenueCat to check whether you own A Smarter
Remote Pro. RevenueCat receives an anonymous app user ID, your
purchase and subscription history, and basic device and app
details such as OS version, app version, and region. It does
not receive your name, email, or payment details.
</div>
<div class="pb-2">
<div class="-ml-6">(c) Voice input:</div> If you use voice
search, your speech is converted to text by Apple&apos;s
speech recognition. Apple may process it on its servers. The
resulting text is sent only to your TV.
</div>
<div class="pb-2">
<div class="-ml-6">(d) Sharing:</div> If you share a TV with
someone, its name, address, and platform are put into a QR
code on your screen for them to scan. It goes from one phone
to the other and does not pass through our servers.
</div>
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">2.</span> How We Use It
</div>
<div class="pb-2 pl-4">
Data about your TVs is used only to control them. Purchase
records are used only to unlock Pro. We do not use any of it for
advertising, and we do not sell it.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">3.</span> Local Network Access
</div>
<div class="pb-2 pl-4">
iOS asks for permission before the app can use your local
network. The app needs it to find and control your TVs. If you
decline, the app cannot reach your TVs. You can change this in
Settings &gt; Privacy &amp; Security &gt; Local Network.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">4.</span> Deleting Your Data
</div>
<div class="pb-2 pl-4">
Removing a TV in the app deletes its stored details and
credentials from your phone. Deleting the app removes all of
them. Subscriptions are managed and cancelled in your App Store
account settings.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">5.</span> Children&apos;s Privacy
</div>
<div class="pb-2 pl-4">
The app does not knowingly collect personal information from
anyone, including children under 13.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">6.</span> Changes to the Privacy Policy
</div>
<div class="pb-2 pl-4">
We may update this policy periodically. Any changes are posted on
this page, so please review it from time to time.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">7.</span> Contact Us
</div>
<div class="pb-2 pl-4">
If you have any questions about this policy, you can contact us{" "}
<A href="/contact" class="text-blue hover-underline-animation">
here
</A>
.
</div>
</div>
</ol>
</div>
</div>
</>
);
}

View File

@@ -1,141 +0,0 @@
/**
* A Smarter Remote terms of use — `freno.me/smarter-remote/terms`.
*
* Linked from the iOS app's paywall (App Review 3.1.2). Builds on Apple's
* standard EULA and adds the Pro subscription and lifetime purchase terms.
*/
import { A } from "@solidjs/router";
import { PageHead } from "~/components/PageHead";
import { EdgeCacheHeaders } from "~/components/EdgeCacheHeaders";
export default function SmarterRemoteTerms() {
return (
<>
<EdgeCacheHeaders maxAge={300} />
<PageHead
title="A Smarter Remote Terms of Use"
description="Terms of use for A Smarter Remote and its Pro subscription."
/>
<div class="bg-base">
<div class="min-h-screen px-[8vw] py-[8vh]">
<div class="py-4 text-xl">A Smarter Remote&apos;s Terms of Use</div>
<div class="py-2">Last Updated: November 2, 2026</div>
<div class="py-2">
These terms cover your use of the A Smarter Remote app (&apos;the
App&apos;). By using the App you agree to them. The App is licensed
to you under Apple&apos;s{" "}
<a
href="https://www.apple.com/legal/internet-services/itunes/dev/stdeula/"
target="_blank"
rel="noopener noreferrer"
class="text-blue hover-underline-animation"
>
Licensed Application End User License Agreement
</a>
, and the terms below add to it.
</div>
<ol>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">1.</span> A Smarter Remote Pro
</div>
<div class="pl-4">
<div class="pb-2">
<div class="-ml-6">(a) What it includes:</div> The free App
controls one TV. Pro removes that limit and unlocks the
features marked Pro in the App.
</div>
<div class="pb-2">
<div class="-ml-6">(b) Yearly subscription:</div> Payment is
charged to your Apple account when you confirm the purchase,
or when any free trial ends. The subscription renews
automatically at the same price unless you cancel at least 24
hours before the end of the current period. Your account is
charged for renewal within 24 hours before the period ends.
Any unused part of a free trial is forfeited when you buy a
subscription.
</div>
<div class="pb-2">
<div class="-ml-6">(c) Lifetime purchase:</div> A one-time
payment that unlocks Pro with no recurring charge.
</div>
<div class="pb-2">
<div class="-ml-6">(d) Managing and refunds:</div> You can
manage or cancel a subscription in your App Store account
settings. Refunds are handled by Apple under its policies.
Restore Purchases in the App re-unlocks Pro on another device
signed in to the same Apple account.
</div>
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">2.</span> Your TVs
</div>
<div class="pb-2 pl-4">
The App controls TVs through the network features their makers
provide. Which features work depends on your TV&apos;s model and
software, which can change without notice. We do not guarantee
that every feature works on every TV. Only control TVs you own
or have permission to use.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">3.</span> No Warranty
</div>
<div class="pb-2 pl-4">
The App is provided &apos;as is&apos;, without warranties of any
kind, to the extent the law allows. To the extent the law allows,
we are not liable for indirect or consequential damages arising
from your use of the App.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">4.</span> Privacy
</div>
<div class="pb-2 pl-4">
How the App handles data is described in the{" "}
<A
href="/smarter-remote/privacy"
class="text-blue hover-underline-animation"
>
Privacy Policy
</A>
.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">5.</span> Changes to These Terms
</div>
<div class="pb-2 pl-4">
We may update these terms periodically. Any changes are posted on
this page. Continuing to use the App after a change means you
accept the new terms.
</div>
</div>
<div class="py-2">
<div class="pb-2 text-lg">
<span class="-ml-4 pr-2">6.</span> Contact Us
</div>
<div class="pb-2 pl-4">
If you have any questions about these terms, you can contact us{" "}
<A href="/contact" class="text-blue hover-underline-animation">
here
</A>
.
</div>
</div>
</ol>
</div>
</div>
</>
);
}

View File

@@ -3,6 +3,7 @@ import { getCookie, setCookie } from "vinxi/http";
import type { H3Event } from "vinxi/http";
import { t } from "~/server/api/utils";
import { logAuditEvent } from "~/server/audit";
import { resolveTrustedClientIp } from "~/server/trusted-client-ip";
import { env } from "~/env/server";
import {
AUTH_CONFIG,
@@ -328,39 +329,15 @@ async function cleanupExpiredRateLimits(): Promise<void> {
}
/**
* Get client IP address from request headers.
* Only trusts X-Forwarded-For outside of local development (set by the Vercel
* edge network in production; trusted in tests so the header-parsing path is
* exercised). In local development, uses the socket address to prevent header
* spoofing.
* Get client IP address for a security decision (rate limiting, audit).
*
* Thin alias over `resolveTrustedClientIp` (./trusted-client-ip), which owns
* the trust order: the previous leftmost `x-forwarded-for` read was
* client-spoofable in our append-only proxy topology, and `x-real-ip` was
* read unconditionally — both fixed there (FRE-371).
*/
export function getClientIP(event: H3Event): string {
// In production on Vercel, X-Forwarded-For is set by the edge network
// and cannot be spoofed by clients. In dev, ignore it.
if (env.NODE_ENV !== "development") {
const forwarded = getHeaderValue(event, "x-forwarded-for");
if (forwarded) {
return forwarded.split(",")[0].trim();
}
const realIP = getHeaderValue(event, "x-real-ip");
if (realIP) {
return realIP;
}
}
// Fallback: try socket remote address
try {
const nodeReq = event.node.req;
if (nodeReq?.socket?.remoteAddress) {
const addr = nodeReq.socket.remoteAddress;
// Clean up IPv6-mapped IPv4 addresses
return addr.replace(/^::ffff:/, "");
}
} catch {
// socket access failed
}
return "unknown";
return resolveTrustedClientIp(event);
}
/**

View File

@@ -150,29 +150,100 @@ describe("Rate Limiting", () => {
});
describe("getClientIP", () => {
it("should extract IP from x-forwarded-for header", () => {
// FRE-371: the read is spoof-resistant now. In the pinned topology
// (client → Cloudflare → origin proxy → app) both proxies APPEND, so
// x-forwarded-for arrives as `<spoof…>, <true client>, <CF edge>` and the
// leftmost hop is whatever the sender typed. Trust order:
// cf-connecting-ip, then x-forwarded-for rightmost-minus-one, then the
// socket peer — never the leftmost hop, never x-real-ip.
it("takes x-forwarded-for rightmost-minus-one, never the leftmost hop", () => {
const event = createMockEvent({
headers: { "x-forwarded-for": "192.168.1.1, 10.0.0.1" }
headers: {
"x-forwarded-for": "6.6.6.6, 192.168.1.1, 10.0.0.1"
}
});
const ip = getClientIP(event);
expect(ip).toBe("192.168.1.1");
});
it("should extract IP from x-real-ip header", () => {
it("prefers cf-connecting-ip (Cloudflare overwrites it at the edge)", () => {
const event = createMockEvent({
headers: {
"cf-connecting-ip": "203.0.113.7",
"x-forwarded-for": "6.6.6.6, 192.168.1.1, 10.0.0.1"
}
});
const ip = getClientIP(event);
expect(ip).toBe("203.0.113.7");
});
it("a rotating leftmost x-forwarded-for does not change the resolved client", () => {
// The spoof the old read fell for: same Cloudflare-recorded client,
// different sender-chosen prefix per request. The key must be stable,
// or each rotation mints a fresh rate-limit bucket.
const first = getClientIP(
createMockEvent({
headers: {
"cf-connecting-ip": "203.0.113.7",
"x-forwarded-for": "1.1.1.1, 203.0.113.7, 10.0.0.1"
}
})
);
const second = getClientIP(
createMockEvent({
headers: {
"cf-connecting-ip": "203.0.113.7",
"x-forwarded-for": "2.2.2.2, 203.0.113.7, 10.0.0.1"
}
})
);
expect(first).toBe("203.0.113.7");
expect(second).toBe(first);
});
it("ignores x-real-ip entirely (client-settable in this topology)", () => {
const event = createMockEvent({
headers: { "x-real-ip": "192.168.1.2" }
});
const ip = getClientIP(event);
expect(ip).toBe("192.168.1.2");
expect(ip).toBe("unknown");
});
it("should prefer x-forwarded-for over x-real-ip", () => {
it("ignores a client-supplied single-hop x-forwarded-for with no trusted proxy above it", () => {
// One hop only = the sender's own invention; discounting the trusted
// hop leaves nothing to believe.
const event = createMockEvent({
headers: { "x-forwarded-for": "6.6.6.6" }
});
const ip = getClientIP(event);
expect(ip).toBe("unknown");
});
it("a non-loopback socket peer is the client and beats every header", () => {
const event = createMockEvent({
remoteAddress: "198.51.100.23",
headers: {
"x-forwarded-for": "192.168.1.1",
"x-real-ip": "192.168.1.2"
"cf-connecting-ip": "6.6.6.6",
"x-forwarded-for": "6.6.6.6, 192.168.1.1, 10.0.0.1",
"x-real-ip": "6.6.6.6"
}
});
const ip = getClientIP(event);
expect(ip).toBe("198.51.100.23");
});
it("a loopback socket peer lets the forwarded headers speak", () => {
const event = createMockEvent({
remoteAddress: "127.0.0.1",
headers: {
"x-forwarded-for": "6.6.6.6, 192.168.1.1, 10.0.0.1"
}
});
@@ -198,13 +269,23 @@ describe("Rate Limiting", () => {
it("should handle IPv6 addresses", () => {
const event = createMockEvent({
headers: {
"x-forwarded-for": "2001:0db8:85a3:0000:0000:8a2e:0370:7334"
"x-forwarded-for":
"2001:0db8:85a3:0000:0000:8a2e:0370:7334, 10.0.0.1"
}
});
const ip = getClientIP(event);
expect(ip).toBe("2001:0db8:85a3:0000:0000:8a2e:0370:7334");
});
it("rejects a candidate hop that is not an address", () => {
const event = createMockEvent({
headers: { "x-forwarded-for": "not-an-ip, 10.0.0.1" }
});
const ip = getClientIP(event);
expect(ip).toBe("unknown");
});
});
describe("rateLimitLogin", () => {

View File

@@ -14,12 +14,15 @@ export function createMockEvent(options: {
cookies?: Record<string, string>;
method?: string;
url?: string;
/** Socket peer the server observed; omit for a peer-less synthetic event. */
remoteAddress?: string;
}): H3Event {
const {
headers = {},
cookies = {},
method = "POST",
url = "http://localhost:3000/"
url = "http://localhost:3000/",
remoteAddress
} = options;
// Build the cookie header string from the cookies object only
@@ -49,7 +52,10 @@ export function createMockEvent(options: {
headers: headersObj,
node: {
req: {
headers: allHeaders
headers: allHeaders,
...(remoteAddress
? { socket: { remoteAddress } }
: {})
},
res: {
cookies: {}

View File

@@ -0,0 +1,171 @@
import { describe, expect, it } from "bun:test";
import {
resolveTrustedClientIp,
isIpAddress,
isLoopbackAddress,
UNKNOWN_CLIENT_IP,
type ClientIpEventLike
} from "./trusted-client-ip";
/**
* Unit coverage for the security-path client-IP read (FRE-371). The caller
* suites (rate-limit.test.ts, report.test.ts) pin the behaviour through
* getClientIP and the route; this file pins the helper's own edges: the peer
* forms a real socket can present, the APIEvent wrapper, and failing closed.
*/
function event(opts: {
headers?: Record<string, string>;
remoteAddress?: string;
wrapped?: boolean;
}): ClientIpEventLike {
const inner: ClientIpEventLike = {
...(opts.remoteAddress
? { node: { req: { socket: { remoteAddress: opts.remoteAddress } } } }
: {}),
request: { headers: new Headers(opts.headers ?? {}) }
};
// An APIEvent wraps its H3Event as `nativeEvent` (routes/api/* shape).
return opts.wrapped ? { nativeEvent: inner } : inner;
}
describe("resolveTrustedClientIp", () => {
it("trust order: non-loopback peer, then cf-connecting-ip, then rightmost-minus-one x-forwarded-for", () => {
// Peer wins.
expect(
resolveTrustedClientIp(
event({
remoteAddress: "198.51.100.23",
headers: {
"cf-connecting-ip": "203.0.113.7",
"x-forwarded-for": "6.6.6.6, 203.0.113.7, 10.0.0.1"
}
})
)
).toBe("198.51.100.23");
// Loopback peer → Cloudflare's record.
expect(
resolveTrustedClientIp(
event({
remoteAddress: "127.0.0.1",
headers: {
"cf-connecting-ip": "203.0.113.7",
"x-forwarded-for": "6.6.6.6, 203.0.113.7, 10.0.0.1"
}
})
)
).toBe("203.0.113.7");
// No CF header → the hop below the trusted proxy.
expect(
resolveTrustedClientIp(
event({
remoteAddress: "127.0.0.1",
headers: { "x-forwarded-for": "6.6.6.6, 203.0.113.7, 10.0.0.1" }
})
)
).toBe("203.0.113.7");
});
it("reads the wrapped H3Event of an APIEvent-shaped input", () => {
expect(
resolveTrustedClientIp(
event({
wrapped: true,
remoteAddress: "127.0.0.1",
headers: { "cf-connecting-ip": "203.0.113.7" }
})
)
).toBe("203.0.113.7");
});
it("never returns a truncated address", () => {
// The full /128 is the bucket key; a truncated one would pool attackers
// with their neighbours.
const ip = resolveTrustedClientIp(
event({
remoteAddress: "127.0.0.1",
headers: { "cf-connecting-ip": "2001:0db8:85a3:0000:0000:8a2e:0370:7334" }
})
);
expect(ip).toBe("2001:0db8:85a3:0000:0000:8a2e:0370:7334");
});
it("fails closed to the stricter unknown key", () => {
// One hop = the sender's own invention; a non-address hop; nothing at all.
expect(
resolveTrustedClientIp(
event({ headers: { "x-forwarded-for": "6.6.6.6" } })
)
).toBe(UNKNOWN_CLIENT_IP);
expect(
resolveTrustedClientIp(
event({ headers: { "cf-connecting-ip": "not-an-ip" } })
)
).toBe(UNKNOWN_CLIENT_IP);
expect(resolveTrustedClientIp(event({}))).toBe(UNKNOWN_CLIENT_IP);
});
it("recognises the loopback forms a real peer presents", () => {
expect(
resolveTrustedClientIp(
event({
// IPv6-mapped IPv4 loopback from a node socket.
remoteAddress: "::ffff:127.0.0.1",
headers: { "cf-connecting-ip": "203.0.113.7" }
})
)
).toBe("203.0.113.7");
expect(
resolveTrustedClientIp(
event({
remoteAddress: "::1",
headers: { "cf-connecting-ip": "203.0.113.7" }
})
)
).toBe("203.0.113.7");
// A non-loopback peer that merely looks odd is still the client.
expect(
resolveTrustedClientIp(
event({
remoteAddress: "::ffff:203.0.113.7",
headers: { "cf-connecting-ip": "6.6.6.6" }
})
)
).toBe("203.0.113.7");
});
it("never believes x-real-ip", () => {
expect(
resolveTrustedClientIp(
event({ headers: { "x-real-ip": "203.0.113.7" } })
)
).toBe(UNKNOWN_CLIENT_IP);
});
});
describe("address helpers", () => {
it("isIpAddress", () => {
expect(isIpAddress("203.0.113.7")).toBe(true);
expect(isIpAddress("::1")).toBe(true);
expect(isIpAddress("::ffff:203.0.113.7")).toBe(true);
expect(isIpAddress("2001:db8::1")).toBe(true);
expect(isIpAddress("not-an-ip")).toBe(false);
expect(isIpAddress("256.1.1.1")).toBe(false);
expect(isIpAddress("")).toBe(false);
expect(isIpAddress(null)).toBe(false);
expect(isIpAddress(undefined)).toBe(false);
});
it("isLoopbackAddress", () => {
expect(isLoopbackAddress("127.0.0.1")).toBe(true);
expect(isLoopbackAddress("127.255.0.3")).toBe(true);
expect(isLoopbackAddress("::1")).toBe(true);
expect(isLoopbackAddress("::ffff:127.0.0.1")).toBe(true);
expect(isLoopbackAddress("203.0.113.7")).toBe(false);
expect(isLoopbackAddress("::2")).toBe(false);
// Unparseable input is not loopback — fails closed.
expect(isLoopbackAddress("garbage")).toBe(false);
});
});

View File

@@ -0,0 +1,250 @@
/**
* The client address, resolved for security decisions (FRE-371).
*
* This is the ONLY client-IP read allowed to feed a security decision — rate
* limiting, abuse quotas, account lockout, audit trails. It exists separately
* from `request-client.ts`'s `resolveClientIp` (the analytics read) on
* purpose:
*
* - `resolveClientIp` carries a standing rule (FRE-369 review, 1d9287a):
* it never feeds a security decision. Its blast radius is a poisoned
* analytics row; ours is granting or denying access.
* - Rate limiting needs the UN-truncated address; the analytics module only
* ever hands out truncated prefixes.
* - `request-client.ts` ships on FRE-366's branch, which is under review;
* a security fix must not merge-depend on an analytics diff.
*
* The address-parsing primitives below are therefore a deliberate port of the
* reviewed ones, not an import. If the two ever drift, this file's version is
* the one security paths follow.
*
* WHY the old read was spoofable: `security.ts`'s `getClientIP` took the
* LEFTMOST `x-forwarded-for` hop, believing the header "cannot be spoofed".
* In the pinned topology — client → Cloudflare → same-host origin proxy →
* app — both proxies *append*, so the header arrives as
* `<client-supplied spoof…>, <true client>, <Cloudflare edge>` and the
* leftmost hop is exactly the one the sender chose. Rotating
* `X-Forwarded-For: <random>` per request minted a fresh rate-limit bucket
* and unlimited bug-report quota. `x-real-ip` was read unconditionally too;
* nothing in this topology overwrites it, so it is equally client text and is
* no longer read at all.
*
* Trust order (matches the convention FRE-366 landed after its security
* review, re-derived here for security stakes):
*
* 1. The direct socket peer, when it is a real routable address: it IS the
* client (dev, direct hit), and any forwarded header from it is the
* client describing itself — ignored.
* 2. `cf-connecting-ip` — reachable only when the peer is the loopback
* origin proxy (or unreadable, e.g. a synthetic event in tests).
* Cloudflare *overwrites* this header at the edge, so a client-supplied
* value never survives. Must parse as an address.
* 3. `x-forwarded-for` counted from the RIGHT: drop the rightmost hop (the
* Cloudflare edge address the origin proxy appended) and take the hop
* before it. Never the leftmost. Must parse as an address.
* 4. "unknown" — callers give unknown-IP keys stricter shared limits, so
* an unresolvable client fails toward a tighter bucket, never a looser
* one.
*
* Deploy-side precondition this code cannot verify (same as the analytics
* read): the origin must be reachable only through Cloudflare and the
* same-host proxy. If the app is directly internet-reachable, a client can
* set every header read here and no in-app read can recover the truth.
*/
/** Cloudflare's "who actually connected" header: overwritten at the edge. */
const CF_CONNECTING_IP = "cf-connecting-ip";
const X_FORWARDED_FOR = "x-forwarded-for";
/**
* Hops discounted from the right of `x-forwarded-for` before the value is
* taken: the Cloudflare edge address the origin proxy appended. Pinned to 1
* because there is exactly one trusted proxy between Cloudflare and the app;
* raise it only if the topology gains a trusted hop.
*/
const FORWARDED_HOPS_TO_DISCOUNT = 1;
/** Returned when no trustworthy address exists. Callers rate-limit it harder. */
export const UNKNOWN_CLIENT_IP = "unknown";
/**
* The shapes this module reads: an H3Event (`security.ts` callers) or a
* SolidStart APIEvent (`routes/api/*`), which wraps its H3Event as
* `nativeEvent`. Every field is optional because test doubles are minimal;
* anything unreadable simply falls through to the next trust step.
*/
export interface ClientIpEventLike {
request?: { headers?: unknown };
headers?: unknown;
node?: {
req?: {
headers?: Record<string, unknown>;
socket?: { remoteAddress?: string };
};
};
nativeEvent?: ClientIpEventLike;
}
/**
* The client address for a security decision, per the trust order above.
* Never truncated: rate-limit buckets are per-address, and truncating would
* let an attacker share a bucket with the innocents behind their /24.
*/
export function resolveTrustedClientIp(event: ClientIpEventLike): string {
// An APIEvent's socket and (authoritative) headers live on the wrapped
// H3Event; a bare APIEvent-shaped double without one is read as-is.
const source = event.nativeEvent ?? event;
const peer = directPeer(source);
if (peer && isIpAddress(peer) && !isLoopbackAddress(peer)) return peer;
// The peer is the loopback origin proxy (or unreadable): forwarded headers
// may be believed, in the order above.
const connecting = cleanHop(headerValue(source, CF_CONNECTING_IP));
if (isIpAddress(connecting)) return connecting;
const hops = (headerValue(source, X_FORWARDED_FOR) ?? "")
.split(",")
.map(cleanHop)
.filter(Boolean);
const index = hops.length - 1 - FORWARDED_HOPS_TO_DISCOUNT;
if (index >= 0 && isIpAddress(hops[index])) return hops[index]!;
return UNKNOWN_CLIENT_IP;
}
/** The socket peer, IPv6-mapped IPv4 flattened (`::ffff:1.2.3.4` → `1.2.3.4`). */
function directPeer(event: ClientIpEventLike): string | null {
try {
const addr = event.node?.req?.socket?.remoteAddress;
return addr ? addr.replace(/^::ffff:/, "") : null;
} catch {
return null;
}
}
/**
* Header read tolerant of every event shape in play: Fetch-style
* `request.headers.get`, an h3 `event.headers` (Headers or plain object), and
* the raw node request headers. Header names are case-insensitive per HTTP;
* the plain-object fallbacks lowercase to match how node and the test doubles
* store them.
*/
function headerValue(event: ClientIpEventLike, name: string): string | null {
try {
const fromHeaders = (h: unknown): string | null => {
if (!h) return null;
if (typeof (h as Headers).get === "function") {
return (h as Headers).get(name) ?? null;
}
if (typeof h === "object") {
const record = h as Record<string, unknown>;
const value = record[name] ?? record[name.toLowerCase()];
return typeof value === "string" ? value : null;
}
return null;
};
return (
fromHeaders(event.request?.headers) ??
fromHeaders(event.headers) ??
fromHeaders(event.node?.req?.headers)
);
} catch {
return null;
}
}
function cleanHop(value: string | null | undefined): string {
return (value ?? "").trim();
}
// ── Addresses: parse and validate ───────────────────────────────────────────
// Ported from the reviewed `request-client.ts` (FRE-366 @ 1d9287a); see the
// header comment for why this is a copy and not an import.
/** Any address this file understands: a dotted quad or an IPv6 literal. */
export function isIpAddress(value: string | null | undefined): boolean {
return value !== null && value !== undefined && addressBytes(value) !== null;
}
/**
* Whether an address is the loopback the production proxy presents.
*
* Parses rather than string-matches, so the forms a real peer can present in —
* `::ffff:127.0.0.53`, `0:0:0:0:0:0:0:1`, an IPv4-mapped literal — are all
* recognised. An address this does not parse is *not* loopback, which fails
* closed: forwarded headers from an unrecognised peer are ignored rather than
* believed (a non-loopback peer short-circuits before headers are read).
*/
export function isLoopbackAddress(ip: string): boolean {
const bytes = addressBytes(ip);
if (!bytes) return false;
if (bytes.length === 4) return bytes[0] === 127;
// IPv6 loopback ::1 — and ::ffff:127/8, which is loopback too.
const mapped = ipv4FromBytes(bytes);
if (mapped) return isLoopbackAddress(mapped);
const tail = bytes[15]!;
return bytes.slice(0, 15).every((byte) => byte === 0) && tail === 1;
}
function isIpv4(value: string): boolean {
const parts = value.split(".");
if (parts.length !== 4) return false;
return parts.every(
(part) => /^\d{1,3}$/.test(part) && Number(part) <= 255 && !/^0\d/.test(part)
);
}
/** 4 bytes for IPv4, 16 for IPv6 (including `::ffff:a.b.c.d`), else null. */
function addressBytes(value: string): number[] | null {
const raw = value.trim().replace(/^\[|\]$/g, "").split("%")[0]!;
if (isIpv4(raw)) return raw.split(".").map(Number);
if (!raw.includes(":")) return null;
const halves = raw.split("::");
if (halves.length > 2) return null;
const compressed = halves.length === 2;
const head = groupsToBytes(halves[0] ?? "");
const tail = groupsToBytes(halves[1] ?? "");
if (!head || !tail) return null;
const total = head.length + tail.length;
if (compressed) {
if (total > 8) return null;
return [...head, ...new Array(8 - total).fill(0), ...tail].flatMap(split16);
}
if (total !== 8) return null;
return [...head, ...tail].flatMap(split16);
}
/** One side of `::` as 16-bit groups; a trailing `1.2.3.4` counts as two. */
function groupsToBytes(side: string): number[] | null {
if (side === "") return [];
const out: number[] = [];
for (const part of side.split(":")) {
if (/^[0-9a-fA-F]{1,4}$/.test(part)) {
out.push(parseInt(part, 16));
} else if (isIpv4(part)) {
const o = part.split(".").map(Number);
out.push((o[0]! << 8) | o[1]!, (o[2]! << 8) | o[3]!);
} else {
return null;
}
}
return out;
}
function split16(group: number): number[] {
return [(group >> 8) & 0xff, group & 0xff];
}
/** The embedded IPv4 of an IPv4-mapped IPv6 address (`::ffff:1.2.3.4`), else null. */
function ipv4FromBytes(bytes: number[]): string | null {
const prefix = bytes.slice(0, 10).every((byte) => byte === 0);
const marker = bytes[10] === 0xff && bytes[11] === 0xff;
if (!prefix || !marker) return null;
return `${bytes[12]}.${bytes[13]}.${bytes[14]}.${bytes[15]}`;
}