feat(plugin): scenario-keyed MEMORY.md index via auto-adopt (v0.12.0)

INDEX_LINE in adopt.js grows from 3 lines to 9 — adds a scenario→tool
table (改 X 影响面 / 谁调用 X / 看 X 源码 / Y 模块长啥样 / 概念查询).
Existing adopted projects refresh once on next SessionStart via
needsRefresh(); new projects get it from first adopt.

Closes the always-loaded context gap: Claude Code knew the 7+5 tool
names but not the natural-language triggers, so sessions silently slid
to Grep/Read when a code-graph tool would be more precise.

See CHANGELOG.md for migration + opt-out details.

Verification: adopt.test.js 37/37 pass; cargo +1.95.0 clippy
--no-default-features and --all-targets clean with -D warnings;
routing_bench 19/20 unchanged (release does not touch ToolRegistry).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: sdsrss

.claude-plugin/marketplace.json

@@ -5,14 +5,14 @@
   },
   "metadata": {
     "description": "AST knowledge graph plugin for Claude Code — semantic search, call graph, HTTP tracing, impact analysis",
-    "version": "0.11.6"
+    "version": "0.12.0"
   },
   "plugins": [
     {
       "name": "code-graph-mcp",
       "source": "./claude-plugin",
       "description": "AST knowledge graph for intelligent code navigation — auto-indexes your codebase and provides semantic search, call graph traversal, HTTP route tracing, and impact analysis via MCP tools",
-      "version": "0.11.6",
+      "version": "0.12.0",
       "author": {
         "name": "sdsrs"
       },

CHANGELOG.md

@@ -1,5 +1,43 @@
 # Changelog
 
+## v0.12.0 — Scenario-keyed MEMORY.md index (auto-adopt template refresh)
+
+Auto-adopt (`claude-plugin/scripts/adopt.js`) now seeds MEMORY.md's sentinel
+block with a 5-row scenario→tool table in addition to the existing tool-name
+list. The always-loaded context gap this closes: Claude Code knew the 7+5 tool
+names but not the natural-language triggers ("who calls X?", "改 X 影响面")
+that should route to them, so sessions silently slid to `Grep` / `Read` when a
+code-graph tool would be more precise. The scenario phrases now live in the
+200-line-capped MEMORY.md itself, not a second-hop `plugin_code_graph_mcp.md`.
+
+### What changes
+
+Sentinel `<!-- code-graph-mcp:begin v1 -->...<!-- code-graph-mcp:end -->` grows
+from 3 lines to 9. Added block (nested under the existing index entry):
+
+    - 场景速查（优先于 Grep）：
+      - 改 X 影响面 → `get_ast_node symbol=X include_impact=true`（或 CLI `code-graph-mcp impact X`）
+      - 谁调用 X / X 被谁用 → `get_call_graph X` 或 `find_references X`
+      - 看 X 源码 / 签名 → `get_ast_node symbol=X`
+      - Y 模块长啥样 → `module_overview` 或 CLI `code-graph-mcp overview Y/`
+      - 概念查询（不知精确名）→ `semantic_code_search "Z"`；字面匹配用 Grep
+
+### Migration — existing adopted projects
+
+`needsRefresh()` detects INDEX_LINE drift automatically; the sentinel block
+rewrites once on next SessionStart. No user action required.
+
+### Opt-out
+
+- Lock current MEMORY.md block against this refresh: `CODE_GRAPH_NO_TEMPLATE_REFRESH=1` (shipped in v0.11.0)
+- Disable auto-adopt entirely for new projects: `CODE_GRAPH_NO_AUTO_ADOPT=1` (shipped in v0.9.0)
+- Downgrade: reinstall `0.11.6` to restore the 3-line INDEX_LINE
+
+### Verification
+
+- `adopt.test.js`: 37/37 green — tests reference the `INDEX_LINE` constant, so the content extension is transparent.
+- `routing_bench`: 19/20 = 95.0% on `anthropic/claude-sonnet-4.5` via OpenRouter — unchanged from v0.11.6. This release doesn't touch `ToolRegistry` descriptions, which is what the bench measures; the adopted MEMORY.md lives outside the oracle's prompt.
+
 ## v0.11.6 — Tool-description tightening (+5% routing P@1) + OpenRouter backend
 
 First run of the routing-recall benchmark landed v0.11.4 at **P@1 = 18/20 = 90.0%**

Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "code-graph-mcp"
-version = "0.11.6"
+version = "0.12.0"
 edition = "2021"
 
 [features]

Cargo.toml

@@ -4,7 +4,7 @@
   "author": {
     "name": "sdsrs"
   },
-  "version": "0.11.6",
+  "version": "0.12.0",
   "keywords": [
     "code-graph",
     "ast",

claude-plugin/.claude-plugin/plugin.json

@@ -13,6 +13,12 @@ const INDEX_LINE = [
   '- [code-graph-mcp](plugin_code_graph_mcp.md) — v0.10.0 起 tools/list 默认 7 核心 + 5 隐藏可调（省启动 token）',
   '  - 核心 7（默认暴露）：`get_call_graph`/`module_overview`/`semantic_code_search`/`ast_search`/`find_references`/`get_ast_node`/`project_map`',
   '  - 进阶 5（隐藏按名可调）：`impact_analysis`/`trace_http_chain`/`dependency_graph`/`find_similar_code`/`find_dead_code`',
+  '  - 场景速查（优先于 Grep）：',
+  '    - 改 X 影响面 → `get_ast_node symbol=X include_impact=true`（或 CLI `code-graph-mcp impact X`）',
+  '    - 谁调用 X / X 被谁用 → `get_call_graph X` 或 `find_references X`',
+  '    - 看 X 源码 / 签名 → `get_ast_node symbol=X`',
+  '    - Y 模块长啥样 → `module_overview` 或 CLI `code-graph-mcp overview Y/`',
+  '    - 概念查询（不知精确名）→ `semantic_code_search "Z"`；字面匹配用 Grep',
 ].join('\n');
 const TEMPLATE_PATH = path.resolve(__dirname, '..', 'templates', 'plugin_code_graph_mcp.md');
 const TARGET_NAME = 'plugin_code_graph_mcp.md';

claude-plugin/scripts/adopt.js

@@ -1,6 +1,6 @@
 {
   "name": "@sdsrs/code-graph-darwin-arm64",
-  "version": "0.11.6",
+  "version": "0.12.0",
   "description": "code-graph-mcp binary for macOS ARM64",
   "license": "MIT",
   "repository": {

npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sdsrs/code-graph-darwin-x64",
-  "version": "0.11.6",
+  "version": "0.12.0",
   "description": "code-graph-mcp binary for macOS x64",
   "license": "MIT",
   "repository": {

npm/darwin-x64/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sdsrs/code-graph-linux-arm64",
-  "version": "0.11.6",
+  "version": "0.12.0",
   "description": "code-graph-mcp binary for Linux ARM64",
   "license": "MIT",
   "repository": {

npm/linux-arm64/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sdsrs/code-graph-linux-x64",
-  "version": "0.11.6",
+  "version": "0.12.0",
   "description": "code-graph-mcp binary for Linux x64",
   "license": "MIT",
   "repository": {