token-pilot 0.45.1 → 0.46.1

package/.claude-plugin/marketplace.json

@@ -6,14 +6,14 @@
   },
   "metadata": {
     "description": "Token Pilot — save 60-90% tokens when AI reads code",
-    "version": "0.45.1"
+    "version": "0.46.1"
   },
   "plugins": [
     {
       "name": "token-pilot",
       "source": "./",
       "description": "Reduces token consumption by 60-90% via AST-aware lazy file reading, structural symbol navigation, and cross-session tool-usage analytics. 23 MCP tools + 25 subagents + budget watchdog hooks.",
-      "version": "0.45.1",
+      "version": "0.46.1",
       "author": {
         "name": "Digital-Threads"
       },

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "token-pilot",
-  "version": "0.45.1",
+  "version": "0.46.1",
   "description": "Saves 60-90% tokens on AI code reading. AST-aware lazy reads, symbol navigation, find_usages, structural git diff/log, edit-safety guard, Task-routing matcher, cross-session telemetry (errors + diagnostics), 25 tp-* subagents tiered to haiku/sonnet/opus with budget watchdog.",
   "author": {
     "name": "Digital-Threads",

package/CHANGELOG.md

@@ -5,6 +5,55 @@ All notable changes to Token Pilot will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.46.1] - 2026-06-18
+
+### Fixed — node:test (`node --test`) TAP output parsing
+
+`test_summary` did not recognise `node --test`: `detectRunner` had no `node`
+case, so node:test output fell through to the generic parser. node:test emits a
+TAP footer (`# pass N` / `# fail N`, number after the word) and `ok N - name`
+points (`ok` before the number), which the generic `<N> passed` regex never
+matches — a green 2/2 run was reported as 0 passed. Added a `node` runner:
+detected by command (`node --test` / `node:test`) or by the TAP footer, parsing
+pass/fail/skipped/tests from the footer with a fallback to counting `ok` /
+`not ok` lines; failure names come from `not ok N - name`. Purely additive — no
+existing parser touched.
+
+### Changed — dev-dependency security + decoupled registry publish (no shipped change)
+
+- Bump `vitest` / `@vitest/coverage-v8` to 4.x — clears the 6 remaining
+  dev-only high advisories in the vitest/vite/esbuild chain. `npm audit` now
+  reports **0 vulnerabilities** (dev + prod). No runtime/package change: dev
+  deps are not shipped to npm consumers; the full 1402-test suite is green on
+  vitest 4.
+- `publish-mcp.yml`: the MCP Registry job no longer hard-depends on npm-job
+  **success** (`if: !cancelled()`). A failed npm publish (e.g. EOTP on a
+  manual-token release) no longer blocks the registry update; re-run the job
+  via `workflow_dispatch` after a manual `npm publish`.
+
+## [0.46.0] - 2026-06-13
+
+### Added — UserPromptSubmit per-turn reinforcement (caveman-style awareness)
+
+The `SessionStart` reminder injects the full mandatory-tool ruleset exactly
+once (start / `/clear` / `/compact`). Over a long conversation that block decays
+out of the model's attention and competing instructions crowd it out, so
+sessions drift back to raw `Read` / `Grep` and stop using token-pilot — even
+with hooks and CLAUDE.md rules in place. The caveman plugin solves the identical
+problem with a `UserPromptSubmit` hook that re-injects a tiny anchor on every
+user message; we now do the same.
+
+New `hook-user-prompt` (`UserPromptSubmit`) emits a one-line **minimal anchor**
+(~30 tokens) on every prompt — the floor that keeps token-pilot in the working
+set. Deliberately a single short line: the heavy full ruleset stays in
+`SessionStart`, so this per-turn channel never undercuts the tool's own token
+budget (no event-log reads, no per-turn growth).
+
+`additionalContext` only — never blocks the prompt. Safe-runner wrapped (always
+exits 0). Respects `sessionStart.enabled`, `TOKEN_PILOT_BYPASS=1`, and a
+dedicated `TOKEN_PILOT_PROMPT_REMINDER=0` opt-out. Wired into `hooks/hooks.json`,
+the `install-hook` installer, and the typo-guard command allowlist.
+
 ## [0.45.1] - 2026-06-11
 
 ### Fixed — refuse a multi-repo workspace parent (cross-project index bleed)

package/README.md

@@ -2,7 +2,7 @@
 
 **Token-efficient AI coding, enforced.** Cuts context consumption in AI coding assistants by up to **90%** without changing the way you work.
 
-> **Why it matters more now:** as frontier models move up in price — Claude's Fable 5 is the most capable (and most expensive-per-token) tier yet — the tokens you *don't* spend reading code are worth more, not less. The savings are in tokens; the value is in tokens × price. Token Pilot keeps the expensive main thread lean so the premium model spends its budget on reasoning, not on re-reading files.
+> **Why it matters more now:** as frontier models move up in price, the tokens you *don't* spend reading code are worth more, not less. The savings are in tokens; the value is in tokens × price. Token Pilot keeps the expensive main thread lean so the premium model spends its budget on reasoning, not on re-reading files.
 
 Three layers, each useful on its own, stronger together:
 

package/agents/tp-api-surface-tracker.md

@@ -9,7 +9,7 @@ tools:
   - mcp__token-pilot__read_symbol
   - Bash
 model: haiku
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: dd184501203fa7f3c73f419c4ffbe33c4be75400cb64a7a51733a3fe23f6e085
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-audit-scanner.md

@@ -11,7 +11,7 @@ tools:
   - Grep
   - Read
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: d172f600bf32277ea6eb4cbbee4542ddd698a986dcd96997d33930561964569b
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-commit-writer.md

@@ -8,7 +8,7 @@ tools:
   - mcp__token-pilot__test_summary
   - mcp__token-pilot__outline
   - Bash
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: de64a406b5176de19f7422619c7de7949b1f28865f225402c9cea9255f377428
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-context-engineer.md

@@ -13,7 +13,7 @@ tools:
   - Edit
   - Glob
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 68b32af2dacd82ebe52c4eec93edb903d452688274c3065218270627c564d8b0
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-dead-code-finder.md

@@ -11,7 +11,7 @@ tools:
   - Grep
   - Read
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: d9b7f5b7ae6f4ae21305c775361bcab097cc774370a6d976c093571d46d55021
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-debugger.md

@@ -12,7 +12,7 @@ tools:
   - Read
   - Bash
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 052413de8d92377edcde6ae5c823f5378db304baccfa29e8866467f42553a500
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-dep-health.md

@@ -9,7 +9,7 @@ tools:
   - Bash
   - Read
 model: haiku
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: e14dc57493d816f8c2e017963e2ef5f66bea50fd0b805a80e8a0d97c968427e7
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-doc-writer.md

@@ -13,7 +13,7 @@ tools:
   - Edit
   - Glob
 model: haiku
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 57d741794ab40e31a7ac49c68ea39a9088f5827cdef866ce81bfca1b7c9180cf
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-history-explorer.md

@@ -10,7 +10,7 @@ tools:
   - Bash
   - Read
 model: haiku
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 7b70fa76a60e3c58a1de4f56c32c0f166424137e203a0cf1c8654e7c9235d904
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-impact-analyzer.md

@@ -12,7 +12,7 @@ tools:
   - mcp__token-pilot__read_symbols
   - Read
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 351a987e11eba63852f5431a16d8eb53104f4f689f82fdcc5a2bf4db948ba92f
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-incident-timeline.md

@@ -8,7 +8,7 @@ tools:
   - mcp__token-pilot__read_symbol
   - Bash
 model: inherit
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: de5722bfea374eaab096c1ae635c37879e7a91370ee3cd0532f4240be03c91eb
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-incremental-builder.md

@@ -13,7 +13,7 @@ tools:
   - Edit
   - Bash
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 375a824d0d847bb5453ec594c7a62ad566ee7e4d92717b0473f771f1a0477c60
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-migration-scout.md

@@ -11,7 +11,7 @@ tools:
   - Grep
   - Glob
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 0334de1bf99b431b65359637d125cda7c44c6f780eb92c57cc538715b1939536
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-onboard.md

@@ -10,7 +10,7 @@ tools:
   - mcp__token-pilot__smart_read
   - mcp__token-pilot__smart_read_many
   - mcp__token-pilot__read_section
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 832e95633fbc8e9b0c10f3e540a327d4be062fb4b3f17a6cce6be13f414e2927
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-performance-profiler.md

@@ -11,7 +11,7 @@ tools:
   - Bash
   - Read
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: b61f06380d80798fa2e49d37bcba0653495bee04dd6bdbc1feff9a75607b0508
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-pr-reviewer.md

@@ -11,7 +11,7 @@ tools:
   - mcp__token-pilot__read_for_edit
   - Read
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: f83f50d05b4f70285ae7afed2b1a406fc436df56e61a0aedbfb31edc7f2b6e66
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-refactor-planner.md

@@ -8,7 +8,7 @@ tools:
   - mcp__token-pilot__outline
   - mcp__token-pilot__read_symbol
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: c5f6fc122c89e16e5cf774045f92169ee3468555320b898171ba13eca5323550
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-review-impact.md

@@ -9,7 +9,7 @@ tools:
   - mcp__token-pilot__module_info
   - Bash
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 8ef3c3341cbfed4eb8dd130126a9683edc57e378c92ff0ca764d584fd941c55c
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-run.md

@@ -16,7 +16,7 @@ tools:
   - Glob
   - Bash
 model: haiku
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 2b08618d34a61f00aafccbda9fed6d83243296dedb83440edbd2d5c28bb6dbc4
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-session-restorer.md

@@ -9,7 +9,7 @@ tools:
   - mcp__token-pilot__session_budget
   - Bash
   - Read
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 529374ed728f5eed5b758b3be3da65624783c0bf0c1a253d7d661a843eb5f767
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-ship-coordinator.md

@@ -11,7 +11,7 @@ tools:
   - Read
   - Grep
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: a60f6ae110eb3138064bce074e8ba26fa0ce5f4659df1624a9d9d3646803391b
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-spec-writer.md

@@ -9,7 +9,7 @@ tools:
   - Read
   - Write
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: c7a4e8b39228fd5158528f389c924c5ff2d98c4b9b05ee0106d54a26c5dc1350
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-test-coverage-gapper.md

@@ -10,7 +10,7 @@ tools:
   - mcp__token-pilot__test_summary
   - Glob
   - Grep
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: be81eed53a3720d146cf89e4a14a7a56577633f7c84c234c412ab70d64c05b11
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-test-triage.md

@@ -8,7 +8,7 @@ tools:
   - mcp__token-pilot__find_usages
   - mcp__token-pilot__read_symbol
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 362ecf4cb03b059421ea26933473700900073dc38b3a7fe271208dfb1ae14f90
 requiredMcpServers:
   - "token-pilot"

package/agents/tp-test-writer.md

@@ -13,7 +13,7 @@ tools:
   - Edit
   - Bash
 model: sonnet
-token_pilot_version: "0.45.1"
+token_pilot_version: "0.46.1"
 token_pilot_body_hash: 269f2fe22ff4517c277d3f56ca67d8a5527b93290ab21079a83ba7af22c1b5a9
 requiredMcpServers:
   - "token-pilot"

package/dist/cli/typo-guard.d.ts

@@ -17,7 +17,7 @@
  * Everything else passes through untouched — a real project root like
  * `/home/user/my-project` or `./subdir` goes to startServer as before.
  */
-export declare const KNOWN_COMMANDS: readonly ["hook-read", "hook-edit", "hook-pre-bash", "hook-pre-grep", "hook-pre-task", "hook-post-bash", "hook-post-task", "hook-session-start", "hook-bootstrap", "hook-subagent-stop", "install-statusline", "install-hook", "uninstall-hook", "install-ast-index", "doctor", "bless-agents", "unbless-agents", "install-agents", "uninstall-agents", "stats", "tool-audit", "save-doc", "list-docs", "init", "migrate-hooks", "errors", "workflow", "--version", "-v", "--help", "-h"];
+export declare const KNOWN_COMMANDS: readonly ["hook-read", "hook-edit", "hook-pre-bash", "hook-pre-grep", "hook-pre-task", "hook-post-bash", "hook-post-task", "hook-session-start", "hook-bootstrap", "hook-subagent-stop", "hook-user-prompt", "install-statusline", "install-hook", "uninstall-hook", "install-ast-index", "doctor", "bless-agents", "unbless-agents", "install-agents", "uninstall-agents", "stats", "tool-audit", "save-doc", "list-docs", "init", "migrate-hooks", "errors", "workflow", "--version", "-v", "--help", "-h"];
 export interface TypoGuardResult {
     kind: "pass-through" | "typo";
     suggestion?: string;

package/dist/cli/typo-guard.js

@@ -37,6 +37,8 @@ export const KNOWN_COMMANDS = [
     "hook-bootstrap",
     // v0.40.0 — canonical subagent-completion capture
     "hook-subagent-stop",
+    // UserPromptSubmit per-turn reinforcement (caveman-style awareness)
+    "hook-user-prompt",
     // v0.42.0 — one-command statusline badge installer
     "install-statusline",
     "install-hook",

package/dist/core/event-log.d.ts

@@ -56,6 +56,13 @@ export interface HookEvent {
      * Optional — absent when no workflow is active.
      */
     workflow_id?: string;
+    /**
+     * Loom spine link — id of the Loom task this event belongs to, when the
+     * session was launched by Loom (LOOM_TASK_ID set). Lets Loom attribute token
+     * savings to a task exactly. Optional — absent outside Loom, so standalone
+     * token-pilot events stay byte-identical to before.
+     */
+    task_id?: string;
     event: "denied" | "allowed" | "bypass" | "pass-through" | "task" | "diagnostic" | string;
     file: string;
     lines: number;

package/dist/core/event-log.js

@@ -115,8 +115,16 @@ export async function appendEvent(projectRoot, event) {
         const wf = event.workflow_id ??
             process.env.TOKEN_PILOT_WORKFLOW_ID ??
             process.env.CLAUDE_CODE_WORKFLOW_ID ??
+            process.env.LOOM_WORKFLOW_ID ??
             undefined;
-        const tagged = wf ? { ...event, workflow_id: wf } : event;
+        // Loom spine: tag the task id when the session was launched by Loom.
+        // Env-driven so call sites stay unchanged; absent outside Loom → no field.
+        const taskId = event.task_id ?? process.env.LOOM_TASK_ID ?? undefined;
+        let tagged = event;
+        if (wf)
+            tagged = { ...tagged, workflow_id: wf };
+        if (taskId)
+            tagged = { ...tagged, task_id: taskId };
         await ensureLogDir(projectRoot);
         await rotateIfNeeded(projectRoot);
         const line = JSON.stringify(tagged) + "\n";

package/dist/handlers/test-summary.js

@@ -73,6 +73,8 @@ export function detectRunner(command, output) {
         return 'rspec';
     if (cmd.includes('mocha'))
         return 'mocha';
+    if (cmd.includes('node --test') || cmd.includes('node:test'))
+        return 'node';
     // Detect from output
     const lower = output.toLowerCase();
     if (lower.includes('vitest') || lower.includes('vite'))
@@ -85,6 +87,9 @@ export function detectRunner(command, output) {
         return 'phpunit';
     if (lower.includes('--- fail:') || lower.includes('--- pass:') || lower.includes('ok  \t'))
         return 'go';
+    // node:test prints a TAP summary footer: "# tests N" + "# pass N" + "# fail N".
+    if (/^#\s*tests\s+\d+/m.test(output) && /^#\s*pass\s+\d+/m.test(output))
+        return 'node';
     return 'generic';
 }
 // ──────────────────────────────────────────────
@@ -103,6 +108,8 @@ export function parseTestOutput(output, runner) {
             return parseGoTest(output);
         case 'cargo':
             return parseCargoTest(output);
+        case 'node':
+            return parseNodeTest(output);
         default:
             return parseGeneric(output);
     }
@@ -243,6 +250,41 @@ function parseGoTest(output) {
     }
     return result;
 }
+function parseNodeTest(output) {
+    const result = { total: 0, passed: 0, failed: 0, skipped: 0, failures: [] };
+    // node:test (`node --test`) prints a TAP summary footer:
+    //   # tests 2
+    //   # pass 2
+    //   # fail 0
+    //   # skipped 0
+    const num = (re) => {
+        const m = output.match(re);
+        return m ? parseInt(m[1], 10) : null;
+    };
+    const pass = num(/^#\s*pass\s+(\d+)/m);
+    const fail = num(/^#\s*fail\s+(\d+)/m);
+    const skip = num(/^#\s*skipped\s+(\d+)/m);
+    const tests = num(/^#\s*tests\s+(\d+)/m);
+    if (pass !== null || fail !== null) {
+        result.passed = pass ?? 0;
+        result.failed = fail ?? 0;
+        result.skipped = skip ?? 0;
+        result.total = tests ?? result.passed + result.failed + result.skipped;
+    }
+    else {
+        // No footer (truncated output) — count the TAP point lines instead.
+        result.passed = (output.match(/^ok\s+\d+/gm) ?? []).length;
+        result.failed = (output.match(/^not ok\s+\d+/gm) ?? []).length;
+        result.total = result.passed + result.failed + result.skipped;
+    }
+    // Failure names come from the TAP point: "not ok 3 - the test name".
+    const failPattern = /^not ok\s+\d+\s*-\s*(.+)$/gm;
+    let match;
+    while ((match = failPattern.exec(output)) !== null) {
+        result.failures.push({ name: match[1].trim(), error: '' });
+    }
+    return result;
+}
 function parseCargoTest(output) {
     const result = { total: 0, passed: 0, failed: 0, skipped: 0, failures: [] };
     // test result: ok. 5 passed; 0 failed; 1 ignored; 0 measured; 0 filtered out

package/dist/hooks/installer.js

@@ -127,6 +127,14 @@ function createHookConfig(options) {
                     hooks: [hookEntry("hook-session-start", options)],
                 },
             ],
+            // Per-turn reinforcement — re-injects a tiny anchor on every user
+            // message so the mandatory-tool rules don't decay out of attention
+            // over a long session (SessionStart fires only once).
+            UserPromptSubmit: [
+                {
+                    hooks: [hookEntry("hook-user-prompt", options)],
+                },
+            ],
             PostToolUse: [
                 {
                     matcher: "Bash",

package/dist/hooks/user-prompt.d.ts

@@ -0,0 +1,30 @@
+/**
+ * UserPromptSubmit reminder hook — per-turn reinforcement of the
+ * token-pilot mandatory-tool rules.
+ *
+ * Why this exists: `hook-session-start` injects the full ruleset exactly
+ * once (start / `/clear` / `/compact`). Over a long conversation that
+ * block decays out of the model's attention and competing instructions
+ * crowd it out, so sessions drift back to raw Read / Grep. The caveman
+ * plugin solves the identical problem by re-injecting a tiny anchor on
+ * EVERY user message; we do the same — one short line per prompt, the
+ * full heavy ruleset stays in SessionStart.
+ *
+ * Contract: emits `additionalContext` only — never blocks the prompt,
+ * never throws. Pure module: the builder takes (enabled, bypass) and
+ * returns a string or null; the thin `index.ts` case does the IO.
+ */
+/**
+ * The per-turn anchor. Deliberately one line — re-sending the full
+ * mandatory block every turn would burn hundreds of tokens per message
+ * inside a tool whose entire point is saving tokens.
+ */
+export declare const MINIMAL_ANCHOR: string;
+/**
+ * Build the per-turn `additionalContext`, or null when disabled /
+ * bypassed (caller emits nothing).
+ */
+export declare function buildPromptReminder(enabled: boolean, bypass: boolean): string | null;
+/** Wrap the message in the UserPromptSubmit hook output envelope. */
+export declare function formatPromptReminderOutput(message: string): string;
+//# sourceMappingURL=user-prompt.d.ts.map

package/dist/hooks/user-prompt.js

@@ -0,0 +1,44 @@
+/**
+ * UserPromptSubmit reminder hook — per-turn reinforcement of the
+ * token-pilot mandatory-tool rules.
+ *
+ * Why this exists: `hook-session-start` injects the full ruleset exactly
+ * once (start / `/clear` / `/compact`). Over a long conversation that
+ * block decays out of the model's attention and competing instructions
+ * crowd it out, so sessions drift back to raw Read / Grep. The caveman
+ * plugin solves the identical problem by re-injecting a tiny anchor on
+ * EVERY user message; we do the same — one short line per prompt, the
+ * full heavy ruleset stays in SessionStart.
+ *
+ * Contract: emits `additionalContext` only — never blocks the prompt,
+ * never throws. Pure module: the builder takes (enabled, bypass) and
+ * returns a string or null; the thin `index.ts` case does the IO.
+ */
+/**
+ * The per-turn anchor. Deliberately one line — re-sending the full
+ * mandatory block every turn would burn hundreds of tokens per message
+ * inside a tool whose entire point is saving tokens.
+ */
+export const MINIMAL_ANCHOR = "[token-pilot] Before raw Read/Grep/git, use the token-pilot tools: " +
+    "smart_read · read_symbol · find_usages · smart_diff / smart_log. " +
+    "Delegate scoped work to tp-* specialists. " +
+    "Raw Read/Grep only with offset/limit or a narrow regex.";
+/**
+ * Build the per-turn `additionalContext`, or null when disabled /
+ * bypassed (caller emits nothing).
+ */
+export function buildPromptReminder(enabled, bypass) {
+    if (!enabled || bypass)
+        return null;
+    return MINIMAL_ANCHOR;
+}
+/** Wrap the message in the UserPromptSubmit hook output envelope. */
+export function formatPromptReminderOutput(message) {
+    return JSON.stringify({
+        hookSpecificOutput: {
+            hookEventName: "UserPromptSubmit",
+            additionalContext: message,
+        },
+    });
+}
+//# sourceMappingURL=user-prompt.js.map

package/dist/index.js

@@ -47,6 +47,7 @@ import { detectDrift, formatDriftFinding } from "./cli/doctor-drift.js";
 import { handleInstallAgents, maybeEmitStartupReminder, } from "./cli/install-agents.js";
 import { handleUninstallAgents } from "./cli/uninstall-agents.js";
 import { appendEvent, applyRetention, } from "./core/event-log.js";
+import { buildPromptReminder, formatPromptReminderOutput, } from "./hooks/user-prompt.js";
 import { handleStats } from "./cli/stats.js";
 import { handleToolAudit } from "./cli/tool-audit.js";
 import { promptYesNo } from "./cli/install-agents.js";
@@ -387,6 +388,27 @@ export async function main(cliArgs = process.argv.slice(2)) {
             });
             return;
         }
+        case "hook-user-prompt": {
+            // UserPromptSubmit per-turn reinforcement. SessionStart injects the
+            // full ruleset once; this re-injects a tiny anchor on every user
+            // message so token-pilot stays in the model's working set instead of
+            // decaying out of attention (the failure mode caveman fixes the same
+            // way). additionalContext only — never blocks the prompt.
+            await runHookEntryPoint({ hook: "hook-user-prompt" }, async () => {
+                const cfg = await loadConfig(process.cwd());
+                // Reuse the awareness toggle — disabling SessionStart reminders
+                // disables per-turn too. TOKEN_PILOT_PROMPT_REMINDER=0 turns off
+                // just the per-turn channel while keeping SessionStart.
+                const enabled = cfg.sessionStart.enabled &&
+                    process.env.TOKEN_PILOT_PROMPT_REMINDER !== "0";
+                const bypass = process.env.TOKEN_PILOT_BYPASS === "1";
+                const message = buildPromptReminder(enabled, bypass);
+                if (message) {
+                    process.stdout.write(formatPromptReminderOutput(message));
+                }
+            });
+            return;
+        }
         case "install-hook":
             await handleInstallHook(cliArgs[1] || process.cwd());
             return;

package/hooks/hooks.json

@@ -76,6 +76,16 @@
         ]
       }
     ],
+    "UserPromptSubmit": [
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node ${CLAUDE_PLUGIN_ROOT}/dist/index.js hook-user-prompt"
+          }
+        ]
+      }
+    ],
     "PostToolUse": [
       {
         "matcher": "Bash",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "token-pilot",
-  "version": "0.45.1",
+  "version": "0.46.1",
   "description": "Save up to 80% tokens when AI reads code — MCP server for token-efficient code navigation, AST-aware structural reading instead of dumping full files into context window",
   "type": "module",
   "main": "dist/index.js",
@@ -67,15 +67,15 @@
   "mcpName": "io.github.Digital-Threads/token-pilot",
   "license": "MIT",
   "dependencies": {
-    "@modelcontextprotocol/sdk": "^1.12.0",
     "@ast-index/cli": "^3.44.0",
+    "@modelcontextprotocol/sdk": "^1.12.0",
     "chokidar": "^4.0.3"
   },
   "devDependencies": {
-    "@vitest/coverage-v8": "^3.2.4",
     "@types/node": "^22.0.0",
+    "@vitest/coverage-v8": "^4.1.8",
     "typescript": "^5.7.0",
-    "vitest": "^3.0.0"
+    "vitest": "^4.1.8"
   },
   "engines": {
     "node": ">=18.0.0"