@sdsrs/code-graph 0.5.42 → 0.5.43

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

@@ -4,7 +4,7 @@
   "author": {
     "name": "sdsrs"
   },
-  "version": "0.5.42",
+  "version": "0.5.43",
   "keywords": [
     "code-graph",
     "ast",

package/claude-plugin/commands/impact.md

@@ -1,19 +1,9 @@
 ---
-description: Analyze the impact scope of changing a symbol before modifying it
+description: Analyze impact scope before modifying a symbol
 argument-hint: <symbol_name>
 ---
 
-# Impact Analysis
+## Impact Analysis
+!`code-graph-mcp impact $ARGUMENTS 2>/dev/null || echo "Symbol not found or no index. Run: code-graph-mcp incremental-index"`
 
-Analyze what will be affected if you change the given symbol.
-
-## Steps
-
-1. Call `get_call_graph(symbol, "callers", depth=3)` to find all upstream callers
-2. Call `get_call_graph(symbol, "callees", depth=2)` to find downstream dependencies
-3. Check if any callers are HTTP route handlers (look for route-related nodes)
-4. Summarize findings:
-   - **Affected files**: list all unique files containing callers/callees
-   - **Affected routes**: list any HTTP routes that flow through this symbol
-   - **Risk level**: LOW (≤3 callers, no routes), MEDIUM (4-10 callers OR 1-2 routes), HIGH (>10 callers OR ≥3 routes)
-5. Present the analysis before making any modifications
+Present the risk assessment and recommend whether it's safe to proceed.

package/claude-plugin/commands/rebuild.md

@@ -2,5 +2,6 @@
 description: Force a full code-graph index rebuild
 ---
 
-Call the `rebuild_index` tool with `{"confirm": true}` to rebuild the index from scratch.
-Report the number of files indexed and nodes created.
+Run via Bash: `code-graph-mcp incremental-index`
+This updates the index incrementally (only changed files).
+For a full rebuild, delete `.code-graph/` first, then run the MCP server.

package/claude-plugin/commands/status.md

@@ -1,8 +1,5 @@
 ---
-description: Show code-graph index status and embedding progress
+description: Show code-graph index status
 ---
 
-Call the `get_index_status` tool and present the results:
-- files_count, nodes_count, edges_count
-- embedding_status and embedding_progress
-- model_available, is_watching
+!`code-graph-mcp health-check --format json 2>/dev/null || echo '{"error":"No index found"}'`

package/claude-plugin/commands/trace.md

@@ -1,15 +1,12 @@
 ---
-description: Trace a full HTTP request flow from route to data layer
-argument-hint: <route_path>
+description: Trace call flow from a handler or route
+argument-hint: <handler_or_route>
 ---
 
-# HTTP Request Flow Tracing
+## Call Graph (callees)
+!`code-graph-mcp callgraph $ARGUMENTS --direction callees --depth 4 2>/dev/null || echo "Symbol not found or no index."`
 
-Trace the complete execution path of an HTTP request.
+## Call Graph (callers)
+!`code-graph-mcp callgraph $ARGUMENTS --direction callers --depth 2 2>/dev/null`
 
-## Steps
-
-1. Call `trace_http_chain(route, depth=5)` to get the full chain
-2. For each key node in the chain, call `get_ast_node` (by node_id, with `context_lines`) to show the implementation
-3. Map the flow: route → middleware → validation → business logic → data access → response
-4. Highlight any error handling, authentication checks, or database operations
+Map the flow and highlight error handling, auth checks, and data access points.

package/claude-plugin/commands/understand.md

@@ -1,21 +1,12 @@
 ---
-description: Deep dive into a module or file's architecture and relationships
+description: Deep dive into a module or file's architecture
 argument-hint: <file_or_dir_path>
 ---
 
-# Module Deep Dive
+## Module Overview
+!`code-graph-mcp overview $ARGUMENTS 2>/dev/null || echo "No index or no symbols found. Run: code-graph-mcp incremental-index"`
 
-Understand what a module does, its public API, and how it connects to the rest of the codebase.
+## Call Graph (top symbols)
+!`code-graph-mcp search "$ARGUMENTS" --limit 5 2>/dev/null`
 
-## Steps
-
-1. Call `module_overview(path)` to get exports, hot paths, and file structure in one call
-2. Call `dependency_graph(file_path)` for the main file to map imports and dependents
-3. For key functions that need deeper understanding, call `get_call_graph` to trace call chains
-4. Summarize:
-   - **Purpose**: what this module does
-   - **Public API**: exported functions/classes with signatures
-   - **Internal structure**: key internal helpers
-   - **Dependencies**: what it imports
-   - **Dependents**: who imports it
-   - **Hot paths**: most-called functions (by caller count)
+Analyze the above and summarize: purpose, public API, key internal helpers, and hot paths.

package/claude-plugin/hooks/hooks.json

@@ -59,6 +59,19 @@
         "description": "Auto-update code graph index after file edits"
       }
     ],
+    "UserPromptSubmit": [
+      {
+        "matcher": "",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"${CLAUDE_PLUGIN_ROOT}/scripts/user-prompt-context.js\"",
+            "timeout": 4
+          }
+        ],
+        "description": "Inject relevant code-graph context based on user's question"
+      }
+    ],
     "SessionStart": [
       {
         "matcher": "startup",
@@ -69,7 +82,7 @@
             "timeout": 5
           }
         ],
-        "description": "StatusLine self-heal, lifecycle sync, and background update launch at session start"
+        "description": "StatusLine self-heal, lifecycle sync, project map injection at session start"
       }
     ]
   }

package/claude-plugin/scripts/pre-edit-guide.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 'use strict';
 // PreToolUse hook: On FIRST Edit/Write call per session window, remind Claude
-// to check impact_analysis before modifying functions. Fast and non-blocking.
+// to check impact analysis before modifying functions. Fast and non-blocking.
 const fs = require('fs');
 const path = require('path');
 const os = require('os');
@@ -16,6 +16,6 @@ try {
 
 fs.writeFileSync(flag, '');
 process.stdout.write(
-  '[code-graph] Before modifying functions, consider running impact_analysis(symbol_name) ' +
-  'to check blast radius (callers, affected routes, risk level).\n'
+  '[code-graph] Before modifying functions, consider checking blast radius:\n' +
+  '  code-graph-mcp impact <function_name>\n'
 );

package/claude-plugin/scripts/pre-explore-guide.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 'use strict';
 // PreToolUse hook: On FIRST Agent call per session window, suggest
-// code-graph tools for structural code understanding before spawning agents.
+// code-graph CLI commands for structural code understanding before spawning agents.
 const fs = require('fs');
 const path = require('path');
 const os = require('os');
@@ -16,9 +16,9 @@ try {
 
 fs.writeFileSync(flag, '');
 process.stdout.write(
-  '[code-graph] For code structure understanding, try code-graph first (one call vs many):\n' +
-  '  project_map(compact=true) \u2192 full architecture overview\n' +
-  '  module_overview(path, compact=true) \u2192 module structure, exports, hot paths\n' +
-  '  get_call_graph(symbol, compact=true) \u2192 trace call chains\n' +
+  '[code-graph] For code structure understanding, try CLI first (one Bash call vs agent):\n' +
+  '  code-graph-mcp map                  \u2190 full architecture overview\n' +
+  '  code-graph-mcp overview src/module  \u2190 module structure and exports\n' +
+  '  code-graph-mcp callgraph symbol     \u2190 trace call chains\n' +
   'Explore agents remain best for: non-code files, runtime behavior, open-ended investigation.\n'
 );

package/claude-plugin/scripts/pre-glob-guide.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 'use strict';
 // PreToolUse hook: On FIRST Glob call per session window, suggest
-// code-graph tools — but only when exploring project structure,
+// code-graph CLI commands — but only when exploring project structure,
 // NOT finding specific files by name.
 const fs = require('fs');
 const path = require('path');
@@ -32,7 +32,7 @@ try {
 fs.writeFileSync(flag, '');
 process.stdout.write(
   '[code-graph] If exploring project structure (not finding specific files):\n' +
-  '  project_map(compact=true) \u2192 all modules, files, key symbols, dependencies\n' +
-  '  module_overview(path, compact=true) \u2192 files and exports within a module\n' +
+  '  code-graph-mcp map              \u2190 project architecture (modules, deps, entry points)\n' +
+  '  code-graph-mcp overview src/mcp \u2190 module symbols grouped by file and type\n' +
   'Glob remains best for: finding specific files, configs, non-code assets.\n'
 );

package/claude-plugin/scripts/pre-search-guide.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 'use strict';
 // PreToolUse hook: On FIRST Grep call per session window, suggest
-// code-graph tools — but only when the pattern looks like code understanding
+// code-graph CLI commands — but only when the pattern looks like code understanding
 // (function names, module patterns), NOT exact string/constant searches.
 const fs = require('fs');
 const path = require('path');
@@ -31,9 +31,9 @@ try {
 
 fs.writeFileSync(flag, '');
 process.stdout.write(
-  '[code-graph] For understanding code relationships, these tools complement Grep:\n' +
-  '  get_call_graph(symbol, compact=true) \u2192 who calls X / what X calls\n' +
-  '  semantic_code_search(query, compact=true) \u2192 find code by concept\n' +
-  '  module_overview(path, compact=true) \u2192 module exports and structure\n' +
-  'Grep remains best for: exact strings, regex, constants, non-code files.\n'
+  '[code-graph] CLI commands for code understanding (via Bash):\n' +
+  '  code-graph-mcp grep "pattern"     \u2190 AST context grep (match + containing function/class)\n' +
+  '  code-graph-mcp search "concept"   \u2190 semantic search (find code by concept, not exact name)\n' +
+  '  code-graph-mcp callgraph symbol   \u2190 call chain tracing\n' +
+  'Grep remains best for: exact strings, constants, regex, non-code files.\n'
 );

package/claude-plugin/scripts/session-init.js

@@ -1,8 +1,9 @@
 #!/usr/bin/env node
 'use strict';
-const { spawn } = require('child_process');
+const { spawn, execSync } = require('child_process');
 const path = require('path');
 const os = require('os');
+const fs = require('fs');
 const {
   install, update, readManifest, getPluginVersion, checkScopeConflict,
   cleanupDisabledStatusline, isPluginInactive, readJson,
@@ -62,12 +63,43 @@ function runSessionInit() {
 
   const lifecycle = syncLifecycleConfig();
   const autoUpdateLaunched = launchBackgroundAutoUpdate();
-  return { inactive: false, lifecycle, autoUpdateLaunched };
+  const mapInjected = injectProjectMap();
+  return { inactive: false, lifecycle, autoUpdateLaunched, mapInjected };
+}
+
+/**
+ * Inject project_map summary into session context if index exists.
+ * Similar to aider's repo-map — gives Claude project structure upfront.
+ */
+function injectProjectMap() {
+  try {
+    const cwd = process.cwd();
+    const dbPath = path.join(cwd, '.code-graph', 'index.db');
+    if (!fs.existsSync(dbPath)) return false;
+
+    const output = execSync('code-graph-mcp map --compact', {
+      cwd,
+      timeout: 5000,
+      encoding: 'utf8',
+      stdio: ['pipe', 'pipe', 'pipe'],
+    });
+
+    if (output && output.trim()) {
+      process.stdout.write(
+        '[code-graph] Project map (indexed):\n' + output.trim() + '\n'
+      );
+      return true;
+    }
+  } catch {
+    // Index not ready or binary not found — skip silently
+  }
+  return false;
 }
 
 module.exports = {
   launchBackgroundAutoUpdate,
   syncLifecycleConfig,
+  injectProjectMap,
   runSessionInit,
 };
 

package/claude-plugin/scripts/user-prompt-context.js

@@ -0,0 +1,115 @@
+#!/usr/bin/env node
+'use strict';
+// UserPromptSubmit hook: inject relevant code-graph context based on user's question.
+// Only activates when user message references code entities + has understanding intent.
+// This is a CODE INDEX, not a memory store — only inject structural code context.
+const { execSync } = require('child_process');
+const fs = require('fs');
+const path = require('path');
+const os = require('os');
+
+// --- Rate limiting ---
+const flag = path.join(os.tmpdir(), '.code-graph-prompt-ctx');
+const COOLDOWN_MS = 60 * 1000; // 1 minute between injections
+try {
+  const stat = fs.statSync(flag);
+  if (Date.now() - stat.mtimeMs < COOLDOWN_MS) process.exit(0);
+} catch { /* first time */ }
+
+// --- Read user message ---
+let message;
+try {
+  const input = JSON.parse(fs.readFileSync('/dev/stdin', 'utf8'));
+  message = (input && input.message) || '';
+} catch {
+  process.exit(0);
+}
+// Chinese chars are ~3 bytes but 1 char; "看看 fts5_search" is only 16 chars
+if (!message || message.length < 8) process.exit(0);
+
+// --- Check index ---
+const cwd = process.cwd();
+const dbPath = path.join(cwd, '.code-graph', 'index.db');
+if (!fs.existsSync(dbPath)) process.exit(0);
+
+// --- Constants ---
+
+const STOP_WORDS = new Set([
+  'this', 'that', 'with', 'from', 'what', 'when', 'which', 'there',
+  'their', 'these', 'those', 'have', 'been', 'some', 'will', 'would',
+  'could', 'should', 'about', 'after', 'before', 'other', 'every',
+  'where', 'while', 'first', 'under', 'still', 'between', 'without',
+  'being', 'through', 'default', 'function', 'method', 'class',
+]);
+
+// --- Detect intent + entities ---
+
+// Skip non-code prompts (commit, push, simple confirmations, chat, instructions, etc.)
+const trimmed = message.trim();
+if (/^(yes|no|ok|commit|push|y|n|done|thanks|thank you|继续|确认|好的|好|是的|不|可以|行|对|提交|推送|没问题|谢谢|发布|更新|编译|安装|卸载|重启|重连|清理)\s*[.!?。！？]?\s*$/i.test(trimmed)) {
+  process.exit(0);
+}
+// Skip action-only prompts without code entities (修复这些问题, 按优先级实施, etc.)
+if (/^(修复|优化|实施|执行|开始|按|实测|帮我|进入|用|重新)/.test(trimmed) && !/[a-zA-Z_]{4,}/.test(trimmed)) {
+  process.exit(0);
+}
+
+// Extract file paths from message
+const filePaths = (message.match(/(?:src|lib|test|pkg|cmd|internal|app|components?)\/[\w/.-]+/g) || [])
+  .slice(0, 2);
+
+// Extract potential symbol names (camelCase, snake_case, PascalCase, qualified like Foo::bar)
+const symbolCandidates = (message.match(/\b(?:[A-Z]\w*(?:::\w+)+|[a-z]\w*(?:_\w+){1,}|[a-z]\w*(?:[A-Z]\w*)+|[A-Z][a-z]+(?:[A-Z][a-z]+)+)\b/g) || [])
+  .filter(s => s.length > 4)
+  .filter(s => !STOP_WORDS.has(s.toLowerCase()))
+  .slice(0, 3);
+
+// Detect intent keywords (EN + ZH, derived from user's actual prompt history)
+const intentImpact = /(?:impact|影响|修改前|改之前|blast radius|before (?:edit|chang|modif)|risk|风险|改动范围|波及|问题在|bug|干扰|冲突|卡)/i.test(message);
+const intentUnderstand = /(?:how does|怎么工作|怎么实现|怎么做|什么|理解|看看|看一下|了解|分析|explain|understand|架构|architecture|structure|overview|模块|概览|干什么|做什么|工作原理|逻辑|机制|流程|功能|结合度|效率|评估|调研|是什么|有什么|能用不|高效不|达标|起作用|科学|深入思考|源码)/i.test(message);
+const intentCallgraph = /(?:who calls|what calls|调用|call(?:graph|er|ee)|trace|链路|追踪|谁调|被谁调|调了谁|上下游|依赖关系|触发|路径|覆盖|介入)/i.test(message);
+const intentSearch = /(?:where is|在哪|find|search|搜索|找|locate|哪里用|哪里定义|定义在|实现在|处理没|在源码|加不加)/i.test(message);
+
+// Need entities AND intent, or strong entity signal (qualified names like Foo::bar)
+const hasQualifiedSymbol = symbolCandidates.some(s => s.includes('::'));
+const hasIntent = intentImpact || intentUnderstand || intentCallgraph || intentSearch;
+if (!hasIntent && !hasQualifiedSymbol && filePaths.length === 0) {
+  process.exit(0);
+}
+
+// --- Run ONE targeted CLI query ---
+let result = '';
+try {
+  if (intentImpact && symbolCandidates.length > 0) {
+    result = run(`code-graph-mcp impact "${symbolCandidates[0]}"`);
+  } else if (filePaths.length > 0 && intentUnderstand) {
+    // Overview of the mentioned file's directory
+    const dir = filePaths[0].replace(/\/[^/]+$/, '/');
+    result = run(`code-graph-mcp overview "${dir}"`);
+  } else if (intentCallgraph && symbolCandidates.length > 0) {
+    result = run(`code-graph-mcp callgraph "${symbolCandidates[0]}" --depth 2`);
+  } else if ((intentSearch || hasQualifiedSymbol) && symbolCandidates.length > 0) {
+    result = run(`code-graph-mcp search "${symbolCandidates[0]}" --limit 8`);
+  } else if (filePaths.length > 0) {
+    const dir = filePaths[0].replace(/\/[^/]+$/, '/');
+    result = run(`code-graph-mcp overview "${dir}"`);
+  }
+} catch {
+  process.exit(0);
+}
+
+if (result && result.trim()) {
+  fs.writeFileSync(flag, ''); // update cooldown
+  process.stdout.write(result.trim() + '\n');
+}
+
+// --- Helpers ---
+
+function run(cmd) {
+  return execSync(cmd, {
+    cwd,
+    timeout: 3000,
+    encoding: 'utf8',
+    stdio: ['pipe', 'pipe', 'pipe'],
+  });
+}

package/claude-plugin/skills/code-navigation.md

@@ -1,35 +1,20 @@
 ---
 name: code-navigation
-description: PROACTIVE code-graph tool selection. Triggers automatically when you need to explore, understand, trace, or modify code. Use BEFORE choosing Grep/Read for code understanding tasks — code-graph tools save 5-20x tokens. Especially critical before modifying functions (impact_analysis first).
+description: Code search and understanding via CLI. Use when exploring code structure, searching by concept, or checking impact before edits.
 ---
 
-# Code Navigation Rules
+# Code Graph CLI
 
-This project has a code-graph MCP server. The MCP instructions (injected at session start) are the authoritative decision table.
+Indexed project. Use Bash — one command replaces multi-file Grep/Read:
 
-**Key principle: code-graph tools SUPERSEDE Grep/Agent for code understanding. Use compact=true for browsing, full when you need signatures or will edit.**
+| Task | Command | Replaces |
+|------|---------|----------|
+| grep + AST context | `code-graph-mcp grep "pattern" [path]` | Grep |
+| search by concept | `code-graph-mcp search "query"` | Grep (no exact name needed) |
+| structural search | `code-graph-mcp ast-search "q" --type fn --returns Result` | — |
+| project map | `code-graph-mcp map` | Read multiple files |
+| module overview | `code-graph-mcp overview src/path/` | Read directory files |
+| call graph | `code-graph-mcp callgraph symbol` | Grep + Read tracing |
+| impact analysis | `code-graph-mcp impact symbol` | — |
 
-## Quick Reference
-
-| Task | Tool | Savings |
-|---|---|---|
-| Architecture overview | `project_map(compact=true)` | 20x vs Read multiple |
-| Who calls X / what X calls | `get_call_graph(symbol, compact=true)` | 13x vs Grep+Read |
-| Understand module/directory | `module_overview(path, compact=true)` | 20x vs Read files |
-| **Before modifying a function** | `impact_analysis(symbol)` FIRST | prevents breakage |
-| Find code by concept | `semantic_code_search(query, compact=true)` | 10x vs Grep |
-| Trace HTTP request flow | `trace_http_chain(route)` | 10x vs Read |
-| Symbol signature+relations | `get_ast_node(node_id)` | 10x vs Read file |
-| File dependencies | `dependency_graph(file)` | 5x vs Grep |
-
-## Workflow Patterns
-
-1. **Quick lookup**: `semantic_code_search(compact=true)` → `get_ast_node(node_id=N)`
-2. **Before edit**: `impact_analysis(symbol)` → Edit
-3. **Understand**: `project_map(compact=true)` → `module_overview(path, compact=true)` → `get_call_graph(symbol)`
-
-## When to use native tools instead
-
-- **Grep**: exact string match, constants, regex patterns, literal text search
-- **Glob**: find files by name/path pattern
-- **Read**: specific file you already know and need to edit
+Still use Grep for exact strings/constants/regex. Still use Read for files you'll edit.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sdsrs/code-graph",
-  "version": "0.5.42",
+  "version": "0.5.43",
   "description": "MCP server that indexes codebases into an AST knowledge graph with semantic search, call graph traversal, and HTTP route tracing",
   "license": "MIT",
   "repository": {
@@ -33,10 +33,10 @@
     "node": ">=16"
   },
   "optionalDependencies": {
-    "@sdsrs/code-graph-linux-x64": "0.5.42",
-    "@sdsrs/code-graph-linux-arm64": "0.5.42",
-    "@sdsrs/code-graph-darwin-x64": "0.5.42",
-    "@sdsrs/code-graph-darwin-arm64": "0.5.42",
-    "@sdsrs/code-graph-win32-x64": "0.5.42"
+    "@sdsrs/code-graph-linux-x64": "0.5.43",
+    "@sdsrs/code-graph-linux-arm64": "0.5.43",
+    "@sdsrs/code-graph-darwin-x64": "0.5.43",
+    "@sdsrs/code-graph-darwin-arm64": "0.5.43",
+    "@sdsrs/code-graph-win32-x64": "0.5.43"
   }
 }