@hegemonart/get-design-done 1.57.1 → 1.57.3

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@hegemonart/get-design-done",
-  "version": "1.57.1",
-  "description": "A design-quality pipeline for AI coding agents: brief, plan, implement, and verify UI work against your design system.",
+  "version": "1.57.3",
+  "description": "A design-quality pipeline for AI coding agents: brief, explore, plan, design, and verify UI work against your design system.",
   "author": "Hegemon",
   "homepage": "https://github.com/hegemonart/get-design-done",
   "repository": {
@@ -23,10 +23,11 @@
     "sdk/",
     "recipes/",
     "docs/i18n/",
-    "dist/claude-code/",
     "scripts/lib/",
     "scripts/cli/",
     "scripts/install.cjs",
+    "scripts/injection-patterns.cjs",
+    "scripts/bootstrap.cjs",
     "SKILL.md",
     "README.md",
     "CHANGELOG.md",
@@ -84,10 +85,15 @@
     "validate:schemas": "node --experimental-strip-types scripts/validate-schemas.ts",
     "validate:frontmatter": "node --experimental-strip-types scripts/validate-frontmatter.ts agents/",
     "detect:stale-refs": "node scripts/detect-stale-refs.cjs",
+    "validate:feature-counts": "node scripts/check-feature-counts.cjs",
+    "validate:registry-tiers": "node scripts/validate-registry-tiers.cjs",
+    "validate:no-internal-refs": "node scripts/validate-no-internal-refs.cjs",
+    "validate:cache-tiers": "node scripts/check-cache-tiers.cjs",
     "scan:injection": "node scripts/run-injection-scanner-ci.cjs",
     "scan:outbound": "node scripts/scan-outbound-network.cjs",
     "scan:ws-bind": "node scripts/scan-ws-bind.cjs",
     "test:size-budget": "node --test test/suite/agent-size-budget.test.cjs",
+    "validate:skill-surface": "node --test test/suite/skill-surface-sync.test.cjs",
     "release:extract-changelog": "node scripts/extract-changelog-section.cjs",
     "verify:version-sync": "node scripts/verify-version-sync.cjs",
     "typecheck:session-runner": "tsc --noEmit",

package/reference/STATE-TEMPLATE.md

@@ -7,10 +7,7 @@ This is the canonical template for the design pipeline's runtime state file.
 - Every subsequent stage (discover, plan, design, verify) reads `.design/STATE.md` at entry and updates it at completion, per the Write Contract below.
 - `.design/` is gitignored (not distributed with the plugin); only this template ships.
 
-**Distinction from `.planning/STATE.md`:**
-- `.planning/STATE.md` is GSD development state - used by the developers building this plugin.
-- `.design/STATE.md` is pipeline runtime state - used by the pipeline when it runs in a user's project.
-- Keep them strictly separate. Cross-references between them are deferred to Phase 6 per CONTEXT.md.
+`.design/` is the runtime state for the design pipeline as it runs in a user's project. The plugin's own internal development workspace lives outside this directory and never reaches a user install.
 
 ---
 
@@ -47,7 +44,7 @@ skipped_stages: ""
 <decisions>
 <!-- Filled by discover stage. Format: -->
 <!-- D-01: [decision text] (locked | tentative) -->
-<!-- Phase 40 (team mode): an optional attribution suffix records provenance for multi-author merges: -->
+<!-- In team mode, an optional attribution suffix records provenance for multi-author merges: -->
 <!-- D-01: [decision text] (locked | tentative) [author=<git-user> co-author=<gdd-instance-id>] -->
 <!-- The suffix is optional + backward-compatible; see reference/multi-author-model.md. -->
 </decisions>
@@ -59,7 +56,7 @@ skipped_stages: ""
 </must_haves>
 
 <prototyping>
-<!-- Phase 25: appended by sketch-wrap-up / spike-wrap-up + the prototype-gate. -->
+<!-- Appended by sketch-wrap-up / spike-wrap-up + the prototype-gate. -->
 <!-- Three child element types, each on its own line: -->
 <!-- <sketch slug="…" cycle="…" decision="D-XX" status="resolved"/> -->
 <!-- <spike slug="…" cycle="…" decision="D-XX" verdict="yes|no|partial" status="resolved"/> -->
@@ -69,7 +66,7 @@ skipped_stages: ""
 </prototyping>
 
 <quality_gate>
-<!-- Phase 25 (Plan 25-03): written by the quality-gate skill (Stage 4.5). -->
+<!-- Written by the quality-gate skill (Stage 4.5). -->
 <!-- Houses a single most-recent <run/> entry — append-mode would be overkill. -->
 <!-- Format: -->
 <!-- <run started_at="…" completed_at="…" status="pass|fail|timeout|skipped" iteration="N" commands_run="lint,typecheck,test"/> -->
@@ -173,22 +170,22 @@ Discover stage populates with observable behaviors. Verify stage updates status.
 
 ### `<prototyping>`
 
-Phase 25 surface (D-01). A checkpoint log - NOT a stage. Tracks sketch and spike outcomes plus cycle-scoped skip suppressions for the prototype gate.
+A checkpoint log - NOT a stage. Tracks sketch and spike outcomes plus cycle-scoped skip suppressions for the prototype gate.
 
 - `<sketch slug=… cycle=… decision=D-XX status=resolved/>` - written by `sketch-wrap-up` after a sketch resolves into a D-XX decision.
 - `<spike slug=… cycle=… decision=D-XX verdict=yes|no|partial status=resolved/>` - written by `spike-wrap-up` after a spike resolves; `verdict` captures the answer.
-- `<skipped at=… cycle=… reason=…/>` - written by the prototype gate when the user declines to sketch/spike at a firing point. Cycle-scoped suppression (D-02): a `<skipped/>` entry suppresses re-asking for the rest of the named cycle.
+- `<skipped at=… cycle=… reason=…/>` - written by the prototype gate when the user declines to sketch/spike at a firing point. The entry provides cycle-scoped suppression: it suppresses re-asking for the rest of the named cycle.
 
 The block is **optional** - fresh STATE.md files do not carry it. The serializer omits the block entirely when no entries exist; appending the first entry is what materializes the block.
 
 ### `<quality_gate>`
 
-Phase 25 surface (Plan 25-03 / D-06..D-09). Captures the most recent run of the Stage 4.5 quality gate (lint / typecheck / test / visual-regression) between Design and Verify. The block houses a single self-closing `<run/>` element - append-mode is overkill, so each gate completion overwrites the entry.
+Captures the most recent run of the Stage 4.5 quality gate (lint / typecheck / test / visual-regression) between Design and Verify. The block houses a single self-closing `<run/>` element - append-mode is overkill, so each gate completion overwrites the entry.
 
 - `started_at` - ISO 8601 at which the parallel command run entered.
 - `completed_at` - ISO 8601 at which the gate produced its terminal status.
-- `status` - `pass | fail | timeout | skipped`. `pass` clears the verify-entry gate; `fail` blocks; `timeout` warns + proceeds (D-07); `skipped` indicates the detection chain resolved zero commands.
-- `iteration` - non-negative integer fix-loop count (D-08). `1` = single clean pass; `N === max_iters` with `status === 'fail'` = bounded exhaustion.
+- `status` - `pass | fail | timeout | skipped`. `pass` clears the verify-entry gate; `fail` blocks; `timeout` warns + proceeds; `skipped` indicates the detection chain resolved zero commands.
+- `iteration` - non-negative integer fix-loop count. `1` = single clean pass; `N === max_iters` with `status === 'fail'` = bounded exhaustion.
 - `commands_run` - comma-separated names of the commands actually executed in Step 2 (e.g., `lint,typecheck,test`). Empty string when `status === 'skipped'`.
 
 The block is **optional** - fresh STATE.md files do not carry it. The serializer omits the block entirely when `quality_gate === null`; the SKILL writes the first `<run/>` to materialize it.
@@ -245,7 +242,7 @@ Every stage that runs in the pipeline MUST follow this contract when reading and
 
 ---
 
-## Notes for Phase 2 implementors
+## Notes for implementors
 
 - Do not add new top-level XML sections without updating this template.
 - The write contract is non-negotiable - stages that skip the read-at-entry step break resume.

package/reference/audit-scoring.md

@@ -247,7 +247,7 @@ Attach to findings under the Visual Hierarchy pillar that relate to compositiona
 
 ### `i18n_readiness`
 
-Attach to findings under the Accessibility pillar (for WCAG 3.1.1 / 3.1.2 violations) or under the Anti-Pattern Compliance pillar (for hardcoded-string / overflow-at-+40% defects). Emitted by the `agents/design-verifier.md` §i18n probes section (Phase 28-06). See [`./i18n.md`](./i18n.md) §WCAG i18n + §Verifier Integration Spec. Does NOT change pillar weights or scores.
+Attach to findings under the Accessibility pillar (for WCAG 3.1.1 / 3.1.2 violations) or under the Anti-Pattern Compliance pillar (for hardcoded-string / overflow-at-+40% defects). Emitted by the `agents/design-verifier.md` §i18n probes section. See [`./i18n.md`](./i18n.md) §WCAG i18n + §Verifier Integration Spec. Does NOT change pillar weights or scores.
 
 ### `verb_axes` (anti-slop)
 

package/reference/cache-tier-doctrine.md

@@ -0,0 +1,46 @@
+# Cache-Tier Doctrine
+
+The Anthropic prompt cache has a 5-minute TTL and is invalidated by a single byte change in the prefix. Every agent body imports `reference/shared-preamble.md` as the first thing it reads; that import expands to a flat byte block at the top of every agent's prompt. Stable bytes there mean stable cache; churning bytes there mean every agent's cache misses on every spawn.
+
+This doctrine pins the bytes that have to stay stable.
+
+## The four tiers
+
+| Tier | Lifetime | Files | Edit cost |
+|------|----------|-------|-----------|
+| **L0** | Framework-invariant | `reference/meta-rules.md` + `reference/shared-preamble.md` | Invalidates cache for every agent simultaneously. CI gate enforces SHA-256 stability; ratchet via `--rebaseline` only when the framework genuinely changes. |
+| **L1** | Per-agent stable | Agent body itself (role, tools contract, output format) | Invalidates cache for the one agent. Edits are routine. |
+| **L2** | Per-spawn dynamic | `<required_reading>` block + per-invocation prompt | Never caches. Edits are free. |
+| **L3** | Reference-only | `reference/*.md` other than L0 | Loaded per-agent on demand; not part of the cache prefix. Edits do not affect cache. |
+
+## The L0 contract
+
+1. **Two files, no more.** L0 = `reference/meta-rules.md` (framework-invariant rules: required reading, writes protocol, deviation handling, completion markers, context-exhaustion + budget) + `reference/shared-preamble.md` (the aggregator that imports meta-rules and contributes the design-family pillar lists). Any third L0 file requires a doctrine update.
+2. **Byte-stable across cycles.** Section heading text, prose order, ordering of bulleted lists, even whitespace runs; all are essential. An edit that "just rewords a sentence" still invalidates cache for every agent for one session per agent.
+3. **CI gate enforces stability.** `scripts/check-cache-tiers.cjs` computes SHA-256 of each L0 file and compares to `test/fixtures/baselines/l0-hashes.json`. Drift fails the build. A real edit requires `--rebaseline` and a baseline-hash commit alongside the L0 edit.
+4. **Pre-warming exists for legitimate L0 edits.** Run `/gdd:warm-cache` after an L0 edit lands to pre-load the new prefix so the next real spawn is a hit rather than a miss.
+
+## What goes where
+
+- A new design-pillar list shared by 5 design-family skills → `shared-preamble.md` (L0). Costs cache miss; ratchet the L0 baseline.
+- A new validator rule for a single agent → that agent body (L1). No L0 impact.
+- A per-invocation context block; the brief, the must-haves, the user's specific request → `<required_reading>` (L2). No cache implications either way.
+- A reference catalog of heuristics, anti-patterns, WCAG thresholds → `reference/*.md` (L3). No cache implications.
+
+## Verification
+
+- `npm run validate:registry-tiers` - confirms `registry.json` entries' `tier` field is one of L0/L1/L2/L3.
+- `npm run validate:cache-tiers` - confirms L0 file SHA-256 matches the baseline.
+- Both run in CI as part of the default test suite.
+
+## When to ratchet
+
+Ratchet (`--rebaseline`) only when:
+
+1. Framework invariant changes (rare).
+2. A design-pillar list shared by ≥3 skills needs updating.
+3. A factual claim that EVERY agent must know becomes false.
+
+If any of those three is true: edit the L0 file, run `node scripts/check-cache-tiers.cjs --rebaseline`, commit both the L0 edit and the updated baseline. Reviewers see the L0-hash diff and understand the cache cost.
+
+If none of the three is true: the edit belongs in L1 (an agent body), L2 (a per-spawn block), or L3 (a reference doc). Reject the L0 edit; relocate.

package/reference/config-schema.md

@@ -122,7 +122,7 @@ Agents never read config directly - the profile is always injected at spawn time
 
 ## .design/budget.json Schema (Phase 10.1)
 
-Governs the optimization layer introduced in Phase 10.1. Read by every `Agent` tool spawn via `hooks/budget-enforcer.js` (PreToolUse). Bootstrap writes defaults if the file is missing.
+Governs the optimization layer introduced in Phase 10.1. Read by every `Agent` tool spawn via `hooks/budget-enforcer.ts` (PreToolUse). Bootstrap writes defaults if the file is missing.
 
 ## Full Schema
 
@@ -215,11 +215,11 @@ If `.design/budget.json` is missing when any `/gdd:*` command runs, `scripts/boo
 
 ## .design/telemetry/costs.jsonl + .design/agent-metrics.json (Phase 10.1)
 
-Phase 10.1 introduces two measurement artifacts written by `hooks/budget-enforcer.js` (PreToolUse on `Agent` spawns) and `scripts/aggregate-agent-metrics.ts` (detached child of the hook + refresh step of `/gdd:optimize`). Both files live under the gitignored `.design/` directory - they are local session state, not committed.
+Phase 10.1 introduces two measurement artifacts written by `hooks/budget-enforcer.ts` (PreToolUse on `Agent` spawns) and `scripts/aggregate-agent-metrics.ts` (detached child of the hook + refresh step of `/gdd:optimize`). Both files live under the gitignored `.design/` directory - they are local session state, not committed.
 
 ### .design/telemetry/costs.jsonl
 
-Append-only ledger. One JSON object per line. Written by `hooks/budget-enforcer.js` on every PreToolUse decision: spawn allowed, spawn blocked by cap, tier downgraded, cache-hit short-circuit, lazy-gate skip. Consumed by the Phase 11 reflector (`agents/design-reflector.md`) and by `/gdd:optimize`.
+Append-only ledger. One JSON object per line. Written by `hooks/budget-enforcer.ts` on every PreToolUse decision: spawn allowed, spawn blocked by cap, tier downgraded, cache-hit short-circuit, lazy-gate skip. Consumed by the Phase 11 reflector (`agents/design-reflector.md`) and by `/gdd:optimize`.
 
 **Row schema** (verbatim from ROADMAP criterion 7):
 
@@ -292,11 +292,11 @@ Both files sit inside `.design/`, which is already `.gitignore`d at the project
 
 ### Refresh cadence
 
-The aggregator is invoked as a detached child process by `hooks/budget-enforcer.js` after every telemetry row write. It is also invoked directly by `/gdd:optimize` before analysis. There is no cron, no daemon, and no scheduled task - metrics are always at most one spawn stale.
+The aggregator is invoked as a detached child process by `hooks/budget-enforcer.ts` after every telemetry row write. It is also invoked directly by `/gdd:optimize` before analysis. There is no cron, no daemon, and no scheduled task - metrics are always at most one spawn stale.
 
 ## .design/cache-manifest.json Schema (Phase 10.1)
 
-Authored and maintained by `skills/cache-manager/SKILL.md`. Read by `hooks/budget-enforcer.js` (PreToolUse on Agent spawns) for short-circuiting cached spawns per D-05. Layer B of the D-08 two-layer cache. Flat KV shape - keys are SHA-256 hex of the deterministic input-hash, values are entry objects. Schema version 1.
+Authored and maintained by `skills/cache-manager/SKILL.md`. Read by `hooks/budget-enforcer.ts` (PreToolUse on Agent spawns) for short-circuiting cached spawns per D-05. Layer B of the D-08 two-layer cache. Flat KV shape - keys are SHA-256 hex of the deterministic input-hash, values are entry objects. Schema version 1.
 
 ## Full Schema
 
@@ -357,22 +357,22 @@ String - ISO-8601 UTC timestamp equal to `written_at + ttl_seconds`. Precomputed
 - TTL is copied into each entry at write time - **not** recomputed on read. Budget.json changes do not retroactively affect existing entries.
 - `expires_at` = `written_at + ttl_seconds`, computed once, stored in the entry.
 - Stale entries are **not** actively purged (lazy cleanup): they remain in the file until overwritten by a new write with the same key, or pruned manually. A proactive reaper is out of v1 scope.
-- Readers (`hooks/budget-enforcer.js`) check `Date.now() / 1000 > Date.parse(expires_at) / 1000`; if true, return miss.
+- Readers (`hooks/budget-enforcer.ts`) check `Date.now() / 1000 > Date.parse(expires_at) / 1000`; if true, return miss.
 
 ## Read/Write contract
 
 - **Single writer**: `skills/cache-manager/SKILL.md` Phase 4 (`write-result-on-completion`). Other code must not write this file directly.
-- **Multiple readers**: `hooks/budget-enforcer.js`, any orchestrator that runs Phase 2 (`lookup`) for dry-run planning. Readers treat malformed JSON as "manifest absent" (empty cache) - do not throw.
+- **Multiple readers**: `hooks/budget-enforcer.ts`, any orchestrator that runs Phase 2 (`lookup`) for dry-run planning. Readers treat malformed JSON as "manifest absent" (empty cache) - do not throw.
 - **Concurrency**: Claude Code agent spawns are serialized at the hook level (PreToolUse is synchronous). No file locking is needed at v1 scale. Concurrent writes from parallel orchestrators are theoretically possible but exceedingly rare; last-writer-wins is acceptable (cache miss on next read re-populates).
 
 ## Bootstrap behavior
 
-If `.design/cache-manifest.json` is missing when `hooks/budget-enforcer.js` reads it, the hook treats every lookup as a miss and the spawn proceeds normally. No bootstrap action is required - the manifest is created lazily on first successful spawn when `skills/cache-manager/SKILL.md` Phase 4 fires. Unlike `.design/budget.json`, missing cache-manifest.json is the **correct** initial state on a fresh repo.
+If `.design/cache-manifest.json` is missing when `hooks/budget-enforcer.ts` reads it, the hook treats every lookup as a miss and the spawn proceeds normally. No bootstrap action is required - the manifest is created lazily on first successful spawn when `skills/cache-manager/SKILL.md` Phase 4 fires. Unlike `.design/budget.json`, missing cache-manifest.json is the **correct** initial state on a fresh repo.
 
 ## Cross-references
 
 - `skills/cache-manager/SKILL.md` - producer; documents the four-phase contract.
-- `hooks/budget-enforcer.js` (Plan 10.1-01) - reader; short-circuits spawns on hit.
+- `hooks/budget-enforcer.ts` (Plan 10.1-01) - reader; short-circuits spawns on hit.
 - `.design/budget.json` - provides `cache_ttl_seconds` default.
 - `.design/telemetry/costs.jsonl` (Plan 10.1-05) - records `cache_hit: true` rows with zero tokens and zero cost when the short-circuit fires.
 - D-05, D-08, D-09 in `.planning/phases/10.1-optimization-layer-cost-governance/10.1-CONTEXT.md` - decision lineage.

package/reference/i18n.md

@@ -465,7 +465,7 @@ The auditor catches a missed 3.1.2 by scanning for untranslated quotes, proper n
 
 ## Verifier Integration Spec
 
-This section specifies two probes for [agents/design-verifier.md](../agents/design-verifier.md). 28-06 implements; this file owns the spec. Both probes attach to a new `### i18n probes` subsection at the end of Phase 1 - Re-Audit + Category Scoring (after the existing `### Category Scores` subsection), and both classify findings under the orthogonal lens-tag `i18n_readiness` (per D-03 / D-07 - no new pillar, no audit-format change).
+This section specifies two probes for [agents/design-verifier.md](../agents/design-verifier.md). The verifier agent implements; this file owns the spec. Both probes attach to a new `### i18n probes` subsection at the end of Stage 1 - Re-Audit + Category Scoring (after the existing `### Category Scores` subsection), and both classify findings under the orthogonal lens-tag `i18n_readiness` (no new pillar, no audit-format change).
 
 ### Probe 1: Hardcoded-string scan
 

package/reference/intel-schema.md

@@ -8,8 +8,11 @@ Path: `.design/intel/` (gitignored - runtime data only)
 The intel store is a set of flat JSON files (slices) that index the design surface.
 Each slice is an independent file. Agents read slices they need; the updater rewrites only changed slices.
 
-Slices are rebuilt by `scripts/build-intel.cjs` (full initial build) and kept current by the
-`gdd-intel-updater` agent (incremental updates triggered by file changes).
+Slices are rebuilt by `scripts/build-intel.cjs` (full initial build) and refreshed on-demand
+by the `gdd-intel-updater` agent (incremental updates when run by the user or a skill that
+spawns it). The store is **not auto-current**: it reflects the last invocation of build-intel
+or the updater. Skills that depend on fresh intel should invoke the updater first or treat
+slice timestamps as best-effort.
 
 ## Slice Definitions
 
@@ -244,6 +247,38 @@ Cross-reference graph: nodes are files, edges are dependency relationships from
 }
 ```
 
+---
+
+### agent-tiers.json
+
+Runtime-neutral tier index per agent; populated by `gdd-intel-updater` so downstream
+tooling reads tier information without re-parsing every agent's markdown frontmatter.
+Both the model-tier (`opus|sonnet|haiku`) and the equivalent runtime-neutral
+reasoning class (`high|medium|low`) are emitted side-by-side using the locked
+equivalence map (opus↔high, sonnet↔medium, haiku↔low).
+
+```json
+{
+  "generated": "<ISO-8601>",
+  "agents": [
+    {
+      "name": "design-planner",
+      "default_tier": "opus",
+      "reasoning_class": "high"
+    },
+    {
+      "name": "design-verifier-gate",
+      "default_tier": "haiku",
+      "reasoning_class": "low"
+    }
+  ]
+}
+```
+
+Consumers: router, budget enforcer, multi-runtime tier resolver.
+Validation: every agent in `agents/*.md` (except README.md) MUST appear exactly once.
+Written atomically (.tmp + rename) when the updater runs; absent on a fresh checkout.
+
 ## Incremental Update Rules
 
 1. On each run, compare current file `mtime` and `git_hash` against `files.json` entries.

package/reference/meta-rules.md

@@ -14,7 +14,7 @@ When the orchestrator's prompt contains a `<required_reading>` block, you MUST r
 
 Only write files declared in your frontmatter `writes:` list. Agents with `reads-only: true` must never call `Write` or `Edit` on any file. If the task appears to require writing outside your declared scope, stop and return a `<blocker>` in STATE.md rather than expanding your write surface.
 
-If your agent runs in a phase that enforces atomic commits (most do), commit only files in your declared `writes:` list. Use the repo commit convention: `docs(phase-N-P): short imperative description` for documentation-class changes, `feat(phase-N-P): ...` for new capability, `fix(phase-N-P): ...` for bug fixes. Phase and plan numbers come from `.design/STATE.md` `phase:` and the invoking plan's frontmatter.
+If your agent runs in a context that enforces atomic commits (most do), commit only files in your declared `writes:` list. Use the repo's conventional-commit convention: `feat(<scope>): ...` for new capability, `fix(<scope>): ...` for bug fixes, `docs(<scope>): ...` for documentation. The `<scope>` should match a recognisable subsystem in the downstream project (e.g. `feat(verify): ...`, `fix(planner): ...`); do not propagate any internal plugin-development numbering.
 
 ## Deviation Handling
 
@@ -43,7 +43,7 @@ The orchestrator detects failure by reading STATE.md for a `<blocker>`, not by t
 
 ## Context-Exhaustion & Budget Awareness
 
-A PostToolUse hook at `hooks/context-exhaustion.js` watches your tool output for the string `<context-exhaustion>` in your response. If you determine you cannot finish the task in the remaining context, emit:
+A PostToolUse hook at `hooks/context-exhaustion.ts` watches your tool output for the string `<context-exhaustion>` in your response. If you determine you cannot finish the task in the remaining context, emit:
 
 ```xml
 <context-exhaustion>
@@ -54,12 +54,12 @@ resume-hint: <one-sentence instruction for a resumption spawn>
 
 …before your completion marker. The hook captures this into STATE.md so the orchestrator can re-spawn you with a narrower scope. Do not guess when you're near exhaustion - only emit when a concrete obstacle (file too large to read, required diff too wide) forced the call.
 
-A PreToolUse hook at `hooks/budget-enforcer.js` intercepts every `Task` spawn (including the one that invoked you). The hook may:
+A PreToolUse hook at `hooks/budget-enforcer.ts` intercepts every `Task` spawn (including the one that invoked you). The hook may:
 - **Short-circuit** your spawn with a cached result from `.design/cache-manifest.json` (transparent - you never run).
 - **Downgrade** your tier to Haiku at the 80% per-task cap soft-threshold, silently (`auto_downgrade_on_cap: true` in `.design/budget.json`, D-03).
 - **Hard-block** your spawn at the 100% per-task or per-phase cap with an actionable error (D-02).
 
-Implication for you as the agent: **do not assume a specific model tier is live.** Your output must be correct whether you run on Haiku, Sonnet, or Opus. If a task genuinely requires reasoning density beyond Haiku, the `size_budget` + `default-tier` combination should have been set at authoring time so the router routes it correctly - the remedy is a frontmatter update (a Phase 11 reflector proposal), not a mid-run assumption.
+Implication for you as the agent: **do not assume a specific model tier is live.** Your output must be correct whether you run on Haiku, Sonnet, or Opus. If a task genuinely requires reasoning density beyond Haiku, the `size_budget` + `default-tier` combination should have been set at authoring time so the router routes it correctly - the remedy is a frontmatter update via the reflector loop, not a mid-run assumption.
 
 ---
 

package/reference/model-tiers.md

@@ -82,7 +82,7 @@ When the budget-enforcer hook resolves which tier to spawn an agent at, it walks
 
 1. **`.design/budget.json.tier_overrides[{agent_name}]`** - per-project configuration. Wins over frontmatter. Use this knob when the project has budget constraints that differ from the plugin defaults.
 2. **Agent frontmatter `default-tier`** (this file's per-agent map) - plugin default. Set at authoring time.
-3. **Hardcoded fallback `"sonnet"`** - belt-and-suspenders in `hooks/budget-enforcer.js` for agents that somehow lack the frontmatter field. Should never fire in practice after Phase 10.1.
+3. **Hardcoded fallback `"sonnet"`** - belt-and-suspenders in `hooks/budget-enforcer.ts` for agents that somehow lack the frontmatter field. Should never fire in practice after Phase 10.1.
 
 Plus two modifiers that can override the above at spawn time:
 
@@ -108,7 +108,7 @@ Phase 11's reflector encodes these heuristics. This section documents the philos
 
 ## Integration Points
 
-- **`hooks/budget-enforcer.js`** (Plan 10.1-01 Task 04) - implements the precedence chain above. `resolveTier(agent, agentDefaultTier, overrides)` is the concrete function.
+- **`hooks/budget-enforcer.ts`** (Plan 10.1-01 Task 04) - implements the precedence chain above. `resolveTier(agent, agentDefaultTier, overrides)` is the concrete function.
 - **`skills/router/SKILL.md`** (Plan 10.1-01 Task 03) - consults this file's tier map + `reference/model-prices.md` to produce `estimated_cost_usd` pre-spawn.
 - **`skills/optimize/SKILL.md`** (Plan 10.1-04) - advisory command; may suggest tier moves based on telemetry + this doc's heuristics.
 - **`agents/design-reflector.md`** (Phase 11, already merged) - may propose edits to this file + paired frontmatter updates. Proposals land in `.design/reflections/` for user review.