vigiles 6.0.0 → 8.0.0

package/README.md

@@ -1,3 +1,44 @@
+<!--
+  README DIRECTION — read before editing; keep changes aligned.
+  This file is the FRONT DOOR + a marketing asset for someone who already lives
+  in Claude Code / Codex. Optimize for a phone-skimmer.
+
+  1. LEAD WITH BENEFITS, not mechanics or vocabulary. Say what the user GETS
+     (a guard that can't silently fail; a CLAUDE.md that stops lying) before how.
+  1b. NEVER OPEN WITH A NEGATIVE, APOLOGY, OR CAVEAT. A bolded lead-in is the
+     FIRST thing a skimmer reads, so it must be the benefit/on-ramp, never a
+     deficiency or competitor: write "Start in plain markdown", NOT "No
+     TypeScript?". Put the STRONGEST proof (e.g. 2/7→7/7) on its OWN line, never
+     buried mid-paragraph, and END a section on the win, not the caveat (demote
+     trade-offs to a trailing aside). Break run-on em-dash/semicolon chains — a
+     paragraph is ≤ ~3 lines, one idea.
+  1c. LEAD WITH THE CONCRETE PAIN the reader already feels — named in THEIR
+     situation, with the SPECIFIC silent failure, not an abstraction. "You
+     installed plugins and wrote skills — but do they actually work? A skill that
+     never fires, a hook that blocks nothing, a CLAUDE.md full of dead refs" beats
+     "reliability for your harness"; "a library with no tests" is the anchoring
+     analogy. This is NOT a 1b violation: a pain about the READER's situation is a
+     hook, not an apology — 1b bans opening with vigiles's OWN deficiency, a
+     caveat, or a competitor, never the user's pain. SAME FOR THE SUBDOCS: open
+     every guide with the concrete pain, THEN the "what this doc is" line + the
+     README uplink (per docs-quality in CLAUDE.md).
+  2. COMPILED SPECS ARE EASY — reassure, never intimidate. You rarely hand-write
+     a .spec.ts: model-invocable skills (edit-spec / strengthen / test-harness)
+     author + edit it, and hooks auto-compile on save. Present the typed spec as
+     the SMOOTH, AUTO-ENFORCED next step you don't think about — the natural
+     evolution (JS→TS: gradual, strict opt-in), NEVER a wall. Markdown + inline
+     mode work with zero TS for anyone not there yet (progressive adoption).
+  3. THE INSTRUMENTS stay first-class — including Eval (measuring whether a skill
+     actually helps is core, not optional). NOTE: Guard / compiled hooks is PARKED
+     FOR LAUNCH (commented out below; see research/roadmap.md "Launch readiness") —
+     so the live set is Lint/Test/Eval ("three instruments"); re-add Guard post-HN.
+  4. SCANNABLE + SHORT — ~200-line cap; punchy table cells, bullets, runnable
+     blocks; benefits over jargon. Push depth into docs/ and LINK it.
+  5. NO INTERNAL VOCABULARY (moat / measurement-authority / flywheel) and NO
+     research/ links — name the user benefit (see public-vs-internal-docs +
+     readme-brevity in CLAUDE.md).
+-->
+
 <p align="center">
   <img src="logo.png" width="140" alt="vigiles logo" />
 </p>
@@ -5,11 +46,7 @@
 <h1 align="center">vigiles</h1>
 
 <p align="center">
-  <strong>Lint, test &amp; measure the harness your AI agent runs on.</strong>
-</p>
-
-<p align="center">
-  Your CLAUDE.md, hooks, and skills steer the agent — but nothing checks they're <em>true</em>, nothing tests they <em>work</em>, and nothing measures whether they actually <em>help</em>. vigiles does all three.
+  <strong>Make the harness your AI agent runs on reliable.</strong>
 </p>
 
 <p align="center">
@@ -20,57 +57,85 @@
 
 ---
 
-`Agent = Model + Harness`. You'd never ship an app without a linter, a test suite,
-and a benchmark — yet the harness steering your agent runs on vibes. vigiles[^name]
-is the deterministic layer for it — and the only one that can **A/B-test what
-actually moves the needle on your Claude subscription**, not metered API. Three
-things, adopt any:
+**You installed a bunch of plugins and wrote a few skills — but do they actually work?**
+A skill that never fires, a safety hook that blocks nothing, a CLAUDE.md full of dead
+references — your harness fails **silently**, and you find out mid-task.
+
+**It's a library with no tests.**
 
-|                |                                                                                                                                                                                                                                                       |
-| -------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| **🔎 Lint**    | Every file path, script, code symbol, and linter rule your CLAUDE.md cites is checked against reality — so a renamed file or a disabled rule can't silently mislead the agent. **[→](docs/verifying-instruction-files.md)**                           |
-| **🧪 Test**    | Hooks, skills, and subagents are code. vigiles tests they _do their job_ — and almost all of it is **deterministic, no API key**; the real-model evals run on your **Claude subscription**, not metered tokens. **[→](docs/harness-testing.md)**      |
-| **📊 Measure** | Does that skill or plugin actually help — or just add cost? **A/B it on real tasks** and read the bill + correctness, not the vendor's headline. The same engine ranks the hyped ecosystem: **what works vs hype**. **[→](docs/measuring-skills.md)** |
+`Agent = Model + Harness` — the model gets the headlines, the harness is the half you
+own. vigiles[^name] makes it prove itself **before the agent ever runs** — verify, test,
+measure. Mostly free, no API key — real-model checks run on your **Claude Pro/Max
+subscription, not metered tokens**.
 
-Pick the one that hurts today. **Works with Claude Code and Codex**
-([`vigiles/codex`](docs/harnesses.md)), and you can
-[teach it your own harness](docs/authoring-an-adapter.md).
+Three instruments, adopt any:
+
+|             |                                                                                                                                                 |
+| ----------- | ----------------------------------------------------------------------------------------------------------------------------------------------- |
+| **🔎 Lint** | Your CLAUDE.md stops lying — every path, script, symbol & linter rule checked against **reality**. **[→](docs/verifying-instruction-files.md)** |
+| **🧪 Test** | Prove your hooks, skills & subagents do their job — **free, no API key**. **[→](docs/harness-testing.md)**                                      |
+| **📊 Eval** | Know if a skill helps or just costs — **A/B on real tasks**, on your own subscription. **[→](docs/measuring-skills.md)**                        |
+
+<!-- PARKED FOR LAUNCH — Guard / compiled hooks. Re-add this row + the ④ section below post-HN. See research/roadmap.md "Launch readiness".
+| **🛡 Guard** | A safety hook that **can't silently fail open** — write a typed function, get a guard that blocks. **[→](docs/compiled-hooks.md)** |
+-->
+
+<p align="center">
+  <img src="vigiles-demo.gif" width="720" alt="vigiles lint catching a file that moved and a script that was renamed" />
+</p>
+<!-- Regenerate the GIF: `python3 scripts/make-demo-gif.py` (output is verbatim CLI; see scripts/demo.sh for a live asciinema recording). -->
+
+**Two ways in** — pick the pain that's yours:
+
+- **Run agents on your own repo?** Start with **Lint** below, or `npx vigiles init`.
+- **Ship plugins to a marketplace?** **[`vigiles scan`](docs/for-plugin-authors.md)**
+  ranks structural health with no key — see the **[plugin-author guide →](docs/for-plugin-authors.md)**.
+
+**Your agent writes the spec — and you can always eject.** You rarely hand-write a
+`.spec.ts`: skills author it for you, **`init` adopts an existing CLAUDE.md
+non-destructively** (untouched until you compile), and plain markdown + inline
+`<!-- vigiles:enforce -->` comments work with zero TypeScript. **Claude Code and
+Codex** ([`vigiles/codex`](docs/harnesses.md)), or [your own
+harness](docs/authoring-an-adapter.md). _(New here? [the FAQ](docs/faq.md) covers the
+"do I write TypeScript / will it touch my files" questions.)_
 
 ## Quick start
 
 **Paste into Claude Code or Codex:**
 
 ```text
-Set up vigiles in this repo with good defaults (lint + test, non-interactive).
-Verify my CLAUDE.md / AGENTS.md references and show me what's stale, then write
-and run a harness test for one of my hooks or skills. Ask me first before gating
-it in CI, adding a real-model eval, or enforcing strictly (--strict).
+Set up vigiles in this repo: run `npx vigiles init` and accept the defaults. If I
+already have a CLAUDE.md or AGENTS.md, adopt it into a spec and show me which
+references are stale. Then install the dep, compile, and write + run one harness
+test for a hook or skill of mine. Don't enforce a spec-per-file or add a real-model
+eval without asking me first.
 ```
 
+The same prompt works in Codex.
+
 Or do it yourself:
 
 ```bash
 npx vigiles init   # sets up lint + test: spec + harness test + CI + plugin
 ```
 
-It's interactive in a terminal and non-interactive for agents/CI (or with
-`--yes`), so "set up vigiles" from a Claude Code / Codex prompt Just Works — and
-it installs a model-invocable **`test-harness` skill**, so afterward you can just
-tell your agent _"test my skills"_ and it picks the tier and writes the test.
+Interactive in a terminal, non-interactive for agents/CI (or `--yes`).
+
+**You don't hand-write any of this — your agent does.** `init` installs
+model-invocable skills, so a plain-English ask does the work (it edits the source
+and recompiles on save; you never touch it by hand):
+
+- _"test my skills"_ → scaffolds **and runs** a trigger/behaviour test (`test-harness`)
+- _"harden my rules"_ → upgrades prose guidance into enforced linter rules (`strengthen`)
+- _"add a rule to my CLAUDE.md"_ → edits the source and recompiles (`edit-spec`)
 
 <details>
 <summary>What <code>init</code> sets up</summary>
 
-- **Both lint and test** by default; scope with `--lint` / `--test` (one or both).
-- Adds `vigiles` to your `devDependencies`.
-- Installs the Claude Code plugin (skills + hooks) via the marketplace —
-  globally, never vendored into your repo.
-- Wires CI as a `zernie/vigiles@v1` workflow (a composite over the same CLI):
-
-  ```yaml
-  - uses: actions/checkout@v4
-  - uses: zernie/vigiles@v1 # lints by default; posts a sticky PR comment + a `valid` output
-  ```
+- **Both lint and test** by default; scope with `--lint` / `--test`.
+- **Already have a CLAUDE.md / AGENTS.md? `init` adopts it** into a spec faithfully and **non-destructively** — your file is left untouched until you choose to `compile` (and `eject` undoes it).
+- Adds `vigiles` to `devDependencies`; installs the Claude Code plugin (skills + hooks) via the marketplace — globally, never vendored.
+- Wires CI as a `zernie/vigiles@v1` workflow (a composite over the same CLI) that posts a sticky PR comment + a `valid` output.
 
 Prefer to write tests yourself? They can be JS **or** TS
 (`*.harness.{mjs,ts}`) — run them with `npx vigiles test`.
@@ -79,11 +144,10 @@ Prefer to write tests yourself? They can be JS **or** TS
 
 ## ① Lint — your CLAUDE.md lies to your agent
 
-Your CLAUDE.md points the agent at `src/auth/login.ts` and tells it to run
-`npm run check`. But the file moved to `src/auth/session.ts` six commits ago, and
-the script was renamed. The agent trusts the stale claim and acts on fiction.
-
-`npx vigiles lint` resolves every reference against reality:
+**Your CLAUDE.md drifts the moment you refactor.** It points the agent at
+`src/auth/login.ts` and says run `npm run check` — but the file moved six commits
+ago and the script was renamed. The agent trusts the stale claim and acts on
+fiction. `npx vigiles lint` resolves every reference against reality:
 
 ```text
 CLAUDE.md:
@@ -92,31 +156,25 @@ CLAUDE.md:
   ✓ @typescript-eslint/no-floating-promises — exists and enabled in eslint config
 ```
 
-File paths, scripts, and code symbols — plus linter rules across **7 catalogs**
-(the rule exists **and is enabled**). Start with one inline comment, no new files;
-step up to a typed `.spec.ts` (compiled to CLAUDE.md, compiler-grade) when you want
-it — and that same spec is what makes your skills & subagents deterministically
-**testable** (a `result()` contract → an assert, no LLM judge — see ② below), not
-just lint-able. **[Full guide →](docs/verifying-instruction-files.md)**
-
-> **Markdown is prose; a typed spec is a _program_.** Opt in and an agent that
-> leaks, exceeds its effect floor, or hands off mismatched data is a **type
-> error** — your **multi-agent pipeline won't compile if the handoffs don't line
-> up**. Others lint prose after the fact; vigiles is a **compiler for harnesses**.
-> Graduated like TypeScript's `strict` — markdown still works, type-safety is
-> opt-in. **[The moat →](research/typed-spec-moat.md)**
-
-**Same cross-reference, any plugin.** `npx vigiles scan` checks a plugin's
-_contracts_ — every subagent tool, `mcp__server__tool`, `mcp_tool` hook, hook
-event, and script path actually **exists and resolves**, not just parses (valid
-YAML ≠ a tool that's real). A **superset of Anthropic's `claude plugin validate`**,
-no key. **[Audit any plugin →](docs/cli.md#scan-dir)**
+File paths, scripts, code symbols — plus linter rules across **7 linters**
+(ESLint, Ruff, Clippy + four more): each rule exists **and is enabled**.
+
+**Start with one inline comment — no new files.** Step up to a typed `.spec.ts`
+(your agent writes it; it compiles to CLAUDE.md) when you want compiler-grade
+guarantees — and `vigiles eject` hands it back to plain markdown anytime.
+**[Full guide →](docs/verifying-instruction-files.md)**
+
+> **Want more? Bad states can stop compiling.** Opt in deeper and a broken
+> hand-off between agents becomes a build error instead of a runtime surprise —
+> graduated like TypeScript's `strict`, on only when you want it.
+> **[How →](docs/compiled-hooks.md)**
 
 ## ② Test — does your harness do its job?
 
-A hook can be wired wrong. A skill's description can fail to trigger — or hijack
-unrelated prompts. Injected context can never reach the model. All of it passes a
-naive "did it run?" check. vigiles tests the assembled harness for real:
+**You wired the hook — but does it actually block?** A skill's description can fail
+to trigger, or hijack unrelated prompts; injected context can silently never reach
+the model. All of it passes a naive "did it run?" check. vigiles tests the
+assembled harness for real:
 
 ```typescript
 import { runHook } from "vigiles/testing";
@@ -133,18 +191,18 @@ It goes well past _"did it fire?"_:
 
 - **Hooks block** what they must — `runHook`, or the real agent CLI via `runHarnessTest`.
 - **Skills trigger** on the right prompts and stay quiet on the wrong ones — recall _and_ precision (`measureTriggerRate`).
-- **Behaviour is good** — score a skill's output directly, or A/B it on-vs-off for the real lift (`measure` / `runEval`, with significance testing).
+- **Behaviour is good** — score a skill's output, or A/B it on-vs-off for the real lift (`measure` / `runEval`, with significance testing).
 - **Safety holds** — the agent _didn't_ push to the wrong branch or hit a paid API; `interceptTools` catches the attempt so the side effect never happens.
 
 Almost every tier runs with **no model and no API key** — milliseconds, on every
-commit. Only the real-model evals need a model, and they run on your own `claude`
-CLI. **[How it works →](docs/harness-testing.md)**
+commit; only the real-model evals need a model, on your own `claude` CLI.
+**[How it works →](docs/harness-testing.md)**
 
-## ③ Measure — does it actually help, or just cost more?
+## ③ Eval — does it actually help, or just cost more?
 
-A skill claims "65% fewer tokens." A plugin promises "3× faster." Stars and
-vibes — **zero measurement**. vigiles A/Bs the claim on real coding tasks, the
-harness loaded exactly as it ships, and reports the **metric triple**:
+**"65% fewer tokens." "3× faster." Says who?** A skill claims it, a plugin promises
+it — stars and vibes, **zero measurement**. vigiles A/Bs the claim on real coding
+tasks, the harness loaded exactly as it ships, and reports the **metric triple**:
 
 ```typescript
 import { measureArms } from "vigiles/testing";
@@ -155,33 +213,76 @@ const r = await measureArms({
   arms: { baseline: {}, skill: { files: { "SKILL.md": THE_SKILL } } },
   measure: (ctx) => ({ cost: ctx.usage.costUsd, correct: check(ctx) }),
 });
-// → is the bill lower? did correctness survive? (not just "fewer output tokens")
 ```
 
-- **The bill (`costUsd`)** — the honest number; it weights cache ~0.1× and output 1×, so a "saved tokens" headline can't hide behind cheap cache.
-- **The optimization target** — whatever the skill claims to move (output tokens, latency, tool calls). Verified on its own terms.
+- **The bill (`costUsd`)** — weights cache ~0.1× / output 1×, so a "saved tokens" headline can't hide behind cheap cache.
+- **The target** — whatever the skill claims to move (output tokens, latency, tool calls), verified on its own terms.
 - **The blast radius** — correctness, a deterministic 1/0. A token win that breaks the code is **not a win**.
 
-The kicker: every run is **your own `claude` CLI on your Pro/Max subscription** —
-so you can measure on every change, and the same engine powers the **ecosystem
-benchmark** ("we A/B-tested the most-hyped skills — what works vs hype"). That's
-the moat: continuous, sub-affordable measurement nobody metered-by-the-token can
-match. **[Measure a skill →](docs/measuring-skills.md)** · **[Why it's affordable →](docs/eval-architecture.md)**
+**Safe to repeat.** Each real-model run is sandboxed (ephemeral dir, egress blocked
+or allow-listed), and `interceptTools` catches an irreversible external — a push, a
+paid API — as an _attempt_, never running it. **[Safety, sandboxing & FAQ →](docs/safety.md)**
+
+**The eval you can actually afford.** promptfoo / DeepEval hit a metered API and
+bill **per token, every run**. vigiles answers most questions with **no model at
+all**, and runs the rest on your own **Claude Pro/Max subscription — $0 extra**. So
+you can measure on every change. **[Eval a skill →](docs/measuring-skills.md)** · **[Why it's affordable →](docs/eval-architecture.md)**
+
+<!-- PARKED FOR LAUNCH — Guard / compiled hooks. Re-add this whole section (and the table row above) post-HN. See research/roadmap.md "Launch readiness".
+
+## ④ Guard — a safety hook that can't silently fail open
+
+**Your safety hook looks like it blocks — and doesn't.** A guard is your last stop
+before something irreversible, but a hand-written one **fails open** without telling
+you. _(Already write safety hooks? This is the power tool.)_ Write a pure typed
+function instead; vigiles emits the exit code, the JSON, and an AST-backed matcher:
+
+```typescript
+import { defineHook, tool, deny, allow } from "vigiles/hook";
+
+export default defineHook({
+  on: "PreToolUse",
+  match: tool("Bash"),
+  decide: (e) =>
+    e.command.runs("git push", { force: true })
+      ? deny("no force-push to a protected branch")
+      : allow(),
+});
+```
+
+**The proof:** a widely-copied OSS safety hook blocks **2/7** of the disaster
+battery. The compiled rewrite blocks **7/7** — measured, not asserted.
+
+You never hand-write the exit code or JSON field (the usual false confidence), the
+matcher is **AST-backed** (it catches the `cd x && git push -f` a glob misses), and
+the artifact is **stamped** so a later hand-edit is refused.
+
+_Scope: this fixes a hook's logic, not the harness's delivery — a subagent's tool
+calls still bypass any PreToolUse hook
+([#34692](https://github.com/anthropics/claude-code/issues/34692)), so it's a strong
+default, not an unbypassable wall._
+**[Compiled hooks — bug classes + trade-offs →](docs/compiled-hooks.md)**
+
+-->
+
+## FAQ
 
-**The eval you can actually afford.** Most of vigiles needs no model at all; the
-rest drives your subscription:
+- **Isn't this just a markdown linter?** No — it checks whether your instruction file is _true_ (every path/script/symbol/rule exists and is enabled), then tests and measures your harness. A style linter can't do any of that.
+- **Do I have to write TypeScript?** No. Lint your markdown with zero new files; when you want a spec, your agent writes it. It's gradual, like TS's `strict`.
+- **Does it overwrite my files?** No. `init` adopts an existing CLAUDE.md _non-destructively_ — untouched until you `compile`, and `eject` reverses it.
+- **Need an API key?** No for almost everything (free, every commit). Real-model evals run on your Claude Pro/Max subscription — $0 metered tokens.
+- **Non-JS repo?** `npx vigiles lint` verifies your CLAUDE.md with no install (Ruff/Clippy/Pylint/… too).
 
-|                        | Runs on                 | Cost                                        |
-| ---------------------- | ----------------------- | ------------------------------------------- |
-| promptfoo, DeepEval, … | metered API SDK         | billed **per token, every run**             |
-| **vigiles**            | your Claude Pro/Max sub | **$0 extra** — and most tiers need no model |
+**[Full FAQ →](docs/faq.md)**
 
 ## More
 
-- **[Plugin health leaderboard →](docs/cli.md#scan-dir)** — point `scan` at a marketplace (e.g. `wshobson/agents`) and it ranks every plugin by structural health (0–100, A–F), worst issues first — still **no key**. Add `--trigger` for the model-gated column: do the skills actually fire?
-- **[CLI & GitHub Action →](docs/cli.md)** — every command, the Action (inputs / output / versioning), and the Claude Code plugin. The full **[lint rules matrix →](docs/verifying-instruction-files.md#the-validation-rules--the-full-matrix)** lives with the linting guide.
-- **[Skills →](docs/skills.md)** — consumer skills installed as a Claude Code plugin: `/plugin marketplace add zernie/vigiles` then `/plugin install vigiles@vigiles` (or let `vigiles init` do it). The model-invocable ones (`test-harness`, `strengthen`, `edit-spec`) fire on their own — ask _"test my skills"_, _"strengthen my rules"_, or _"add a rule to CLAUDE.md"_ and the agent reaches for them; `adopt-spec` and `linter-docs` are user-invoked.
-- **[Docs index →](docs/README.md)** · **[Research →](research/README.md)** · **[Related tools →](docs/related-tools.md)** (ast-grep, Dependency Cruiser, Ruler, rulesync).
+- **[CLI →](docs/cli.md)** — every command and the plugin · **[GitHub Action →](docs/github-action.md)** — run it in CI. The full **[lint rules matrix →](docs/verifying-instruction-files.md#the-validation-rules--the-full-matrix)** lives with the linting guide.
+- **[Skills →](docs/skills.md)** — the skills `init` installs, and how the model-invocable ones trigger.
+- **[Ship plugins? The plugin-author guide →](docs/for-plugin-authors.md)** — scan a draft for structural health, make your skills fire for users, then rank a whole marketplace (0–100, A–F, worst issues first) — **no key**.
+- **[Docs index →](docs/README.md)** · **[API reference →](https://zernie.github.io/vigiles/)** · **[Related tools →](docs/related-tools.md)** (ast-grep, Dependency Cruiser, Ruler, rulesync).
+- **[Stability →](STABILITY.md)** — 0.x: the CLI is stable; the library API is still evolving; experimental surfaces are marked.
+- **Not for you if** you want a model/capability benchmark or runtime guardrails in the request path — vigiles is build-/CI-time.
 - Companion to [Feedback Loop Is All You Need](https://zernie.com/blog/feedback-loop-is-all-you-need).
 
 ## License

package/dist/action-gate.js

@@ -13,7 +13,7 @@ exports.loadActionGates = loadActionGates;
  *
  * It is the same deterministic gate primitive (reuses `runGate` + the
  * author-time reference resolution), re-anchored from step → action. Delivered
- * as a PostToolUse hook (`vigiles action-hook`): exit 2 blocks the action and
+ * as a PostToolUse hook (`vigiles hook-runtime action`): exit 2 blocks the action and
  * feeds the reason back, exit 0 allows it.
  *
  * Config: `.vigiles/action-gates.json` → `{ "gates": [ { on, gate, when? } ] }`.

package/dist/adapters/claude-code/agent-runtime.d.ts

@@ -10,7 +10,7 @@
  * doesn't list.
  *
  * This is the same emit-a-hook pattern the skill runtime already ships
- * (`src/skill-runtime.ts`): there a `Stop` hook reads the active skill's
+ * (`src/adapters/claude-code/skill-runtime.ts`): there a `Stop` hook reads the active skill's
  * compiled SKILL.md and runs its result gate; here a `PreToolUse` hook reads
  * the active agent's compiled `.md`, parses its `tools:` allowlist, and
  * allows/denies the tool call. The compiled markdown's frontmatter is the
@@ -66,21 +66,56 @@ export interface PreToolDecision {
  *   feeding the contract back to the model so it self-corrects.
  */
 export declare function decidePreToolUse(allowed: readonly string[] | null, tool: string): PreToolDecision;
-/** Record the subagent currently dispatched, so PreToolUse enforces its contract. */
+/**
+ * Read the active-agent stack (oldest → newest; the dispatched subagent chain).
+ * Back-compat: a legacy single-slot `{ agent: string }` marker reads as a one-frame
+ * stack; a malformed file or non-string entries → an empty stack (fail-open).
+ */
+export declare function readActiveStack(cwd: string): string[];
+/**
+ * Push a dispatched subagent onto the active stack (a `PreToolUse` spawn). The
+ * PreToolUse gate then enforces this subagent's contract until it returns. Under
+ * nesting each dispatch pushes a frame, so the chain is tracked, not overwritten.
+ */
+export declare function pushActiveAgent(cwd: string, agentPath: string): void;
+/**
+ * Pop the top frame — the subagent returned (`SubagentStop`), so control returns
+ * to its PARENT (the next frame down), whose contract the gate enforces again.
+ * This is the nesting-safe close, distinct from {@link clearActiveAgent} (which
+ * drops the whole stack). Popping an empty stack is a no-op.
+ */
+export declare function popActiveAgent(cwd: string): void;
+/**
+ * Push a subagent frame (the manual `agent-start` fallback + the deterministic
+ * spawn open-signal both use this). A single call is equivalent to a one-frame
+ * stack, so the top — what the gate reads — is this agent. Alias of
+ * {@link pushActiveAgent} kept under the historical name.
+ */
 export declare function setActiveAgent(cwd: string, agentPath: string): void;
-/** Clear the active-agent marker (the subagent finished). */
+/**
+ * Clear the WHOLE stack (a hard reset / session end). Distinct from
+ * {@link popActiveAgent}, which returns to the parent frame. Idempotent.
+ */
 export declare function clearActiveAgent(cwd: string): void;
-/** The path of the active agent's compiled `.md`, or null when none is active. */
+/**
+ * The active agent's compiled `.md` — the STACK TOP — or null when none is active.
+ * The gate reads the top, so a returned nested subagent reveals its parent's
+ * contract again (the contract-escape fix).
+ */
 export declare function readActiveAgent(cwd: string): string | null;
 /**
- * EXPERIMENTAL — parked (P3), flat-only, do NOT auto-wire. This deterministic
- * subagent-window tracking assumes FLAT dispatch (one active subagent at a time);
- * it is NOT nesting-safe — Claude Code v2.1.172 added depth-5 nested subagents, so
- * correct tracking needs a depth-aware STACK (push on dispatch, pop on SubagentStop)
- * + verifying the spawn tool name (`Agent` vs `Task`). See
- * research/effect-boundary-design.md ("Why dropped").
+ * EXPERIMENTAL — parked (P3), do NOT auto-wire. The subagent-window tracking is
+ * now nesting-safe: a depth-aware STACK (push on dispatch, pop on SubagentStop —
+ * see {@link pushActiveAgent}/{@link popActiveAgent}) closes the contract-escape
+ * the flat single-slot model allowed under Claude Code v2.1.172 depth-5 nesting
+ * (certified in research/prototypes/.../AgentWindowStack.tla). The open signal
+ * recognizes BOTH spawn tool names (`Task` and the nested-spawn `Agent`), gated on
+ * a resolvable `subagent_type` so a non-spawn call never opens a frame. Still
+ * parked because the `effect()` sub-region goal it served was dropped (see
+ * research/effect-boundary-design.md, "Why dropped") — the stack is shipped for
+ * when active-agent contract enforcement under nesting is wanted on its own.
  *
- * Resolve a `Task` tool's `subagent_type` to the compiled agent `.md` to
+ * Resolve a spawn tool's `subagent_type` to the compiled agent `.md` to
  * activate, or null when none is found. The DETERMINISTIC open signal that
  * replaces the model-invoked `agent-start`: Claude Code fires `PreToolUse` for
  * the parent's `Task` dispatch (and `SubagentStop` when it returns), so the

package/dist/adapters/claude-code/agent-runtime.js

@@ -11,7 +11,7 @@
  * doesn't list.
  *
  * This is the same emit-a-hook pattern the skill runtime already ships
- * (`src/skill-runtime.ts`): there a `Stop` hook reads the active skill's
+ * (`src/adapters/claude-code/skill-runtime.ts`): there a `Stop` hook reads the active skill's
  * compiled SKILL.md and runs its result gate; here a `PreToolUse` hook reads
  * the active agent's compiled `.md`, parses its `tools:` allowlist, and
  * allows/denies the tool call. The compiled markdown's frontmatter is the
@@ -28,6 +28,9 @@ exports.parseAgentTools = parseAgentTools;
 exports.parseAgentToolList = parseAgentToolList;
 exports.parseAgentPurity = parseAgentPurity;
 exports.decidePreToolUse = decidePreToolUse;
+exports.readActiveStack = readActiveStack;
+exports.pushActiveAgent = pushActiveAgent;
+exports.popActiveAgent = popActiveAgent;
 exports.setActiveAgent = setActiveAgent;
 exports.clearActiveAgent = clearActiveAgent;
 exports.readActiveAgent = readActiveAgent;
@@ -100,46 +103,114 @@ function decidePreToolUse(allowed, tool) {
     };
 }
 // ---------------------------------------------------------------------------
-// Active-agent tracking (mirrors .vigiles/active-skill.json)
+// Active-agent tracking — a depth-aware STACK (mirrors .vigiles/active-skill.json)
 // ---------------------------------------------------------------------------
+//
+// Claude Code v2.1.172 added nested subagents (a subagent with the spawn tool can
+// dispatch its own, up to depth 5). A single active-agent slot is NOT nesting-safe:
+// when an inner subagent returns, clearing the whole slot drops the OUTER agent's
+// contract while it is still running, so the PreToolUse gate then allows a tool the
+// outer subagent forbids — a CONTRACT ESCAPE. The fix (certified in TLC, see
+// research/prototypes/typed-spec-formal-verification/AgentWindowStack.tla) is a
+// STACK: push on dispatch, pop on SubagentStop (back to the parent), gate on the
+// stack TOP. Counterexample the flat model fails and the stack model passes:
+// Open(writer); Open(writer); Stop; Call(Bash).
 const ACTIVE_PATH = ".vigiles/active-agent.json";
-/** Record the subagent currently dispatched, so PreToolUse enforces its contract. */
-function setActiveAgent(cwd, agentPath) {
+/**
+ * Read the active-agent stack (oldest → newest; the dispatched subagent chain).
+ * Back-compat: a legacy single-slot `{ agent: string }` marker reads as a one-frame
+ * stack; a malformed file or non-string entries → an empty stack (fail-open).
+ */
+function readActiveStack(cwd) {
+    const p = (0, node_path_1.resolve)(cwd, ACTIVE_PATH);
+    if (!(0, node_fs_1.existsSync)(p))
+        return [];
+    try {
+        const parsed = JSON.parse((0, node_fs_1.readFileSync)(p, "utf-8"));
+        if (Array.isArray(parsed.stack)) {
+            return parsed.stack.filter((x) => typeof x === "string");
+        }
+        // legacy single-slot format
+        if (typeof parsed.agent === "string")
+            return [parsed.agent];
+        return [];
+    }
+    catch {
+        return [];
+    }
+}
+function writeActiveStack(cwd, stack) {
     const p = (0, node_path_1.resolve)(cwd, ACTIVE_PATH);
+    if (stack.length === 0) {
+        if ((0, node_fs_1.existsSync)(p))
+            (0, node_fs_1.rmSync)(p);
+        return;
+    }
     (0, node_fs_1.mkdirSync)((0, node_path_1.dirname)(p), { recursive: true });
-    (0, node_fs_1.writeFileSync)(p, JSON.stringify({ agent: agentPath }) + "\n");
+    (0, node_fs_1.writeFileSync)(p, JSON.stringify({ stack }) + "\n");
+}
+/**
+ * Push a dispatched subagent onto the active stack (a `PreToolUse` spawn). The
+ * PreToolUse gate then enforces this subagent's contract until it returns. Under
+ * nesting each dispatch pushes a frame, so the chain is tracked, not overwritten.
+ */
+function pushActiveAgent(cwd, agentPath) {
+    writeActiveStack(cwd, [...readActiveStack(cwd), agentPath]);
+}
+/**
+ * Pop the top frame — the subagent returned (`SubagentStop`), so control returns
+ * to its PARENT (the next frame down), whose contract the gate enforces again.
+ * This is the nesting-safe close, distinct from {@link clearActiveAgent} (which
+ * drops the whole stack). Popping an empty stack is a no-op.
+ */
+function popActiveAgent(cwd) {
+    const stack = readActiveStack(cwd);
+    stack.pop();
+    writeActiveStack(cwd, stack);
 }
-/** Clear the active-agent marker (the subagent finished). */
+/**
+ * Push a subagent frame (the manual `agent-start` fallback + the deterministic
+ * spawn open-signal both use this). A single call is equivalent to a one-frame
+ * stack, so the top — what the gate reads — is this agent. Alias of
+ * {@link pushActiveAgent} kept under the historical name.
+ */
+function setActiveAgent(cwd, agentPath) {
+    pushActiveAgent(cwd, agentPath);
+}
+/**
+ * Clear the WHOLE stack (a hard reset / session end). Distinct from
+ * {@link popActiveAgent}, which returns to the parent frame. Idempotent.
+ */
 function clearActiveAgent(cwd) {
     const p = (0, node_path_1.resolve)(cwd, ACTIVE_PATH);
     if ((0, node_fs_1.existsSync)(p))
         (0, node_fs_1.rmSync)(p);
 }
-/** The path of the active agent's compiled `.md`, or null when none is active. */
+/**
+ * The active agent's compiled `.md` — the STACK TOP — or null when none is active.
+ * The gate reads the top, so a returned nested subagent reveals its parent's
+ * contract again (the contract-escape fix).
+ */
 function readActiveAgent(cwd) {
-    const p = (0, node_path_1.resolve)(cwd, ACTIVE_PATH);
-    if (!(0, node_fs_1.existsSync)(p))
-        return null;
-    try {
-        const parsed = JSON.parse((0, node_fs_1.readFileSync)(p, "utf-8"));
-        return typeof parsed.agent === "string" ? parsed.agent : null;
-    }
-    catch {
-        return null;
-    }
+    const stack = readActiveStack(cwd);
+    return stack.length > 0 ? stack[stack.length - 1] : null;
 }
 // ---------------------------------------------------------------------------
 // PreToolUse-hook decision
 // ---------------------------------------------------------------------------
 /**
- * EXPERIMENTAL — parked (P3), flat-only, do NOT auto-wire. This deterministic
- * subagent-window tracking assumes FLAT dispatch (one active subagent at a time);
- * it is NOT nesting-safe — Claude Code v2.1.172 added depth-5 nested subagents, so
- * correct tracking needs a depth-aware STACK (push on dispatch, pop on SubagentStop)
- * + verifying the spawn tool name (`Agent` vs `Task`). See
- * research/effect-boundary-design.md ("Why dropped").
+ * EXPERIMENTAL — parked (P3), do NOT auto-wire. The subagent-window tracking is
+ * now nesting-safe: a depth-aware STACK (push on dispatch, pop on SubagentStop —
+ * see {@link pushActiveAgent}/{@link popActiveAgent}) closes the contract-escape
+ * the flat single-slot model allowed under Claude Code v2.1.172 depth-5 nesting
+ * (certified in research/prototypes/.../AgentWindowStack.tla). The open signal
+ * recognizes BOTH spawn tool names (`Task` and the nested-spawn `Agent`), gated on
+ * a resolvable `subagent_type` so a non-spawn call never opens a frame. Still
+ * parked because the `effect()` sub-region goal it served was dropped (see
+ * research/effect-boundary-design.md, "Why dropped") — the stack is shipped for
+ * when active-agent contract enforcement under nesting is wanted on its own.
  *
- * Resolve a `Task` tool's `subagent_type` to the compiled agent `.md` to
+ * Resolve a spawn tool's `subagent_type` to the compiled agent `.md` to
  * activate, or null when none is found. The DETERMINISTIC open signal that
  * replaces the model-invoked `agent-start`: Claude Code fires `PreToolUse` for
  * the parent's `Task` dispatch (and `SubagentStop` when it returns), so the

package/dist/adapters/claude-code/effect-region.js

@@ -8,7 +8,7 @@ exports.hasEffectBoundary = hasEffectBoundary;
  * Effect-boundary position-aware state tracking.
  * Mirrors active-unit tracking (.vigiles/active-agent.json / .vigiles/active-skill.json).
  * The PreToolUse hook reads this file to decide whether the agent is inside an
- * effect boundary (set by `vigiles effect-enter`) or outside it.
+ * effect boundary (set by `vigiles hook-runtime effect-enter`) or outside it.
  */
 const node_fs_1 = require("node:fs");
 const node_path_1 = require("node:path");