feat: v0.43.0 — module_route tool, statusline session+MCP savings, CC 2.1.16x

Bundled release capturing everything actionable from ast-index 3.41→3.45
and Claude Code 2.1.163→2.1.167.

Added
- module_route MCP tool (#23) wrapping ast-index 3.44 `module-route`:
  transitive dependency path(s) between two modules, text/json/mermaid/dot
  output. Full-profile only. Handler + client method + validation +
  server dispatch + tool definition + tests.
- statusline Claude.ai rate limits (CC 2.1.80+): `5h:NN% 7d:NN%` from the
  `rate_limits` payload (schema verified against the 2.1.167 bundle).
- effort: low on the guide/install/stats skills (CC 2.1.16x).

Fixed — statusline
- bare [TP] in monorepos/subdirs/worktrees: walk up to the project root
  (nearest ancestor with .token-pilot/), like git finds .git.
- undercounted savings: the badge summed only hook-denial savings
  (hook-events.jsonl). It now also sums MCP-tool savings
  (tool-calls.jsonl, tokensWouldBe − tokensReturned). The MCP server
  stamps each tool-calls row with the real CLAUDE_CODE_SESSION_ID (was
  empty), so both sources split per session.
- frozen-looking number: render session + project with one decimal,
  e.g. [TP s:12.3k · 172.6k].

Changed
- version 0.43.0 (package/plugin/marketplace + help banner 22→23).
- docs: README + docs/tools.md tool count and module_route entry.

Chore
- remove dead imports in src/ast-index/client.ts.

Notes (no code, CC 2.1.16x): additionalContext-on-failed-tool fix,
SubagentStop background_tasks/session_crons, glob deny rules — documented
in README/CHANGELOG. MessageDisplay hook deliberately deferred: output
contract unconfirmable from the minified bundle and it saves no tokens
(display-only).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: shahinyanm

.claude-plugin/marketplace.json

@@ -5,15 +5,15 @@
     "email": "shahinyanm@gmail.com"
   },
   "metadata": {
-    "description": "Token Pilot \u2014 save 60-90% tokens when AI reads code",
-    "version": "0.42.3"
+    "description": "Token Pilot — save 60-90% tokens when AI reads code",
+    "version": "0.43.0"
   },
   "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. 22 MCP tools + 19 subagents + budget watchdog hooks.",
-      "version": "0.42.3",
+      "version": "0.43.0",
       "author": {
         "name": "Digital-Threads"
       },

.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "token-pilot",
-  "version": "0.42.3",
+  "version": "0.43.0",
   "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",
@@ -21,7 +21,9 @@
   "mcpServers": {
     "token-pilot": {
       "command": "sh",
-      "args": ["${CLAUDE_PLUGIN_ROOT}/start.sh"]
+      "args": [
+        "${CLAUDE_PLUGIN_ROOT}/start.sh"
+      ]
     }
   }
 }

CHANGELOG.md

@@ -5,6 +5,107 @@ 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).
 
+## [0.43.0] - 2026-06-06
+
+Bundled release: everything actionable from **ast-index 3.41→3.45** and
+**Claude Code 2.1.163→2.1.167**, plus a statusline fix.
+
+### Added — `module_route` MCP tool (ast-index 3.44)
+
+New tool wrapping ast-index's `module-route` command: the transitive
+dependency path(s) between two modules. Answers "how does module A reach
+module B through the import graph?", traces coupling, and can emit a
+dependency diagram.
+
+- `from` / `to` (required), plus `all`, `maxPaths` (≤200), `maxDepth`
+  (≤50), `viaKind` (`api`/`implementation`/`all`), `format`
+  (`text`/`json`/`mermaid`/`dot`).
+- Machine formats (json/mermaid/dot) pass through clean — no header that
+  would corrupt a diagram/parse. Text format gets a `MODULE ROUTE: a → b`
+  header.
+- Empty output explains both causes (modules unrelated **or** the
+  module-dependency graph isn't indexed — `ast-index rebuild`).
+- Full-profile only (not in nav/edit/minimal sets) — a specialised
+  analysis tool, like the audit tools.
+- `exec` already sets `AST_INDEX_WALK_UP=1`, so the route resolves from a
+  monorepo subdir without passing `--walk-up`.
+
+### Fixed — statusline now counts MCP-tool savings, not just hook denials
+
+The badge summed only `hook-events.jsonl` (`savedTokens` from intercepted
+raw `Read`/`Grep`). The larger share — savings from the MCP tools
+themselves (`smart_read`, `outline`, `find_usages`, …) — lives in
+`tool-calls.jsonl` and was **invisible** to the badge, so the displayed
+figure undercounted real savings (often by an order of magnitude).
+
+Two changes fix it:
+
+- The MCP server now stamps every `tool-calls.jsonl` row with the real
+  Claude Code session id, read from `CLAUDE_CODE_SESSION_ID` (exported to
+  child processes; verified against the 2.1.167 bundle — the same value
+  the hooks receive and the statusline payload carries). Previously these
+  rows had an empty `session_id`, so they could not be attributed to a
+  session at all.
+- The statusline sums **both** logs: `hook-events.jsonl` `savedTokens`
+  plus `tool-calls.jsonl` (`tokensWouldBe − tokensReturned`), each split
+  into the current session and the project total. Zero/negative deltas
+  (pass-throughs) are ignored.
+
+### Added — statusline Claude.ai rate limits (CC 2.1.80+)
+
+The badge now appends `5h:42% 7d:13%` when the statusline payload carries
+`rate_limits` (subscribers only, after the first API response). Schema
+verified against the 2.1.167 bundle: `rate_limits.five_hour.used_percentage`
+and `rate_limits.seven_day.used_percentage`. Unlike the cumulative token
+total, these numbers move every turn — the badge finally shows something
+live. Parsed with `sed` (no `jq` dependency) and whitelisted to digits.
+
+### Fixed — statusline bare `[TP]` in monorepos / subdirs / worktrees
+
+The badge resolved the events log only at the exact `cwd` from the
+payload. In a monorepo (or worktree, or any session `cd`'d into a
+subpackage) the `.token-pilot/` dir is at the repo root, so the lookup
+failed and rendered a bare `[TP]` with no token count. The script now
+**walks up** to the nearest ancestor with `.token-pilot/` (bounded to 40
+levels), the same way git finds `.git`.
+
+### Changed — statusline shows session + project savings (`s:12.3k · 172.6k`)
+
+The badge now renders **both** the current session's saved tokens (the
+number you watch climb during a run) and the cumulative project total,
+e.g. `[TP s:12.3k · 172.6k]`. A fresh session that hasn't saved anything
+yet falls back to the project total alone (never an empty badge after
+first use). Numbers render with one decimal (`172.6k`, not `172k`) so a
+single turn's ~100-token savings is visible on each render — whole-`k`
+rounding previously made the figure look frozen.
+
+### Added — `effort: low` on the bundled skills (CC 2.1.16x)
+
+`guide` / `install` / `stats` declare `effort: low` — they render static
+output or run one CLI call, so they don't need a high-effort model.
+Faster and cheaper when invoked.
+
+### Notes — Claude Code 2.1.16x integration (no code change required)
+
+- **`additionalContext` no longer dropped on a failed tool call** — our
+  PreToolUse routing guidance now survives a failed call (we benefit
+  automatically).
+- **`SubagentStop` input gained `background_tasks` / `session_crons`** —
+  available to our subagent-stop hook for future budget feedback.
+- **Glob deny rules (`"*"` denies all tools)** — usable as a second
+  enforcement layer on top of our deny hooks for stricter TP adoption.
+
+### Deferred — `MessageDisplay` hook (researched, not shipped)
+
+CC's new `MessageDisplay` hook transforms assistant text *as displayed*.
+Deliberately NOT wired this release: (1) the output contract (the field
+that returns replacement text) is not confirmable from the minified
+2.1.167 bundle — and our rule is never to ship an unverified CC field
+beside working hooks; (2) it is display-only, so it saves no input or
+output tokens (the text is already generated and already in context).
+That is caveman's cosmetic-output domain, not token-pilot's input domain.
+Revisit once a live MessageDisplay payload pins the contract.
+
 ## [0.30.0] - 2026-04-19
 
 ### Added — `minimal` profile (5 tools, near-zero overhead)

README.md

@@ -43,7 +43,7 @@ Creates (or merges into) `.mcp.json` with `token-pilot` + [`context-mode`](https
 
 ## What You Get
 
-- **22 MCP tools** — structural reads, symbol search, git analysis, session analytics → [tools reference](docs/tools.md)
+- **23 MCP tools** — structural reads, symbol search, git analysis, module routing, session analytics → [tools reference](docs/tools.md)
 - **PreToolUse hooks** — block heavy `Grep`/`Bash`/`Read` calls; redirect to efficient alternatives → [hooks & modes](docs/hooks.md)
 - **25 `tp-*` subagents** (Claude Code only) — MCP-first delegates with haiku/sonnet model tiers and budget enforcement → [agents reference](docs/agents.md)
 - **Tool profiles** — trim advertised `tools/list` to save ~2 k tokens per session → [profiles & config](docs/configuration.md)

agents/tp-api-surface-tracker.md

@@ -9,7 +9,7 @@ tools:
   - mcp__token-pilot__read_symbol
   - Bash
 model: haiku
-token_pilot_version: "0.42.3"
+token_pilot_version: "0.43.0"
 token_pilot_body_hash: dd184501203fa7f3c73f419c4ffbe33c4be75400cb64a7a51733a3fe23f6e085
 requiredMcpServers:
   - "token-pilot"

agents/tp-audit-scanner.md

@@ -11,7 +11,7 @@ tools:
   - Grep
   - Read
 model: sonnet
-token_pilot_version: "0.42.3"
+token_pilot_version: "0.43.0"
 token_pilot_body_hash: d172f600bf32277ea6eb4cbbee4542ddd698a986dcd96997d33930561964569b
 requiredMcpServers:
   - "token-pilot"

agents/tp-commit-writer.md

@@ -8,7 +8,7 @@ tools:
   - mcp__token-pilot__test_summary
   - mcp__token-pilot__outline
   - Bash
-token_pilot_version: "0.42.3"
+token_pilot_version: "0.43.0"
 token_pilot_body_hash: de64a406b5176de19f7422619c7de7949b1f28865f225402c9cea9255f377428
 requiredMcpServers:
   - "token-pilot"

agents/tp-context-engineer.md

@@ -13,7 +13,7 @@ tools:
   - Edit
   - Glob
 model: sonnet
-token_pilot_version: "0.42.3"
+token_pilot_version: "0.43.0"
 token_pilot_body_hash: 68b32af2dacd82ebe52c4eec93edb903d452688274c3065218270627c564d8b0
 requiredMcpServers:
   - "token-pilot"

agents/tp-dead-code-finder.md

@@ -11,7 +11,7 @@ tools:
   - Grep
   - Read
 model: sonnet
-token_pilot_version: "0.42.3"
+token_pilot_version: "0.43.0"
 token_pilot_body_hash: d9b7f5b7ae6f4ae21305c775361bcab097cc774370a6d976c093571d46d55021
 requiredMcpServers:
   - "token-pilot"

agents/tp-debugger.md

@@ -12,7 +12,7 @@ tools:
   - Read
   - Bash
 model: sonnet
-token_pilot_version: "0.42.3"
+token_pilot_version: "0.43.0"
 token_pilot_body_hash: 052413de8d92377edcde6ae5c823f5378db304baccfa29e8866467f42553a500
 requiredMcpServers:
   - "token-pilot"