chore: prepare 0.5.0 release

CHANGELOG.md

@@ -11,7 +11,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - **Documentation repositioning** — README and getting-started now lead with AI metadata tracking, memory as "level up"
 
-## [0.4.0] - 2026-02-15
+## [0.5.0] - 2026-02-16
 
 ### Added
 
@@ -31,12 +31,22 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - Detects memory type (decision, gotcha, convention, fact) from patterns
   - Infers tags from conventional commit scope and file paths
   - Adds AI-Agent, AI-Model, AI-Confidence, AI-Lifecycle, AI-Memory-Id trailers
+- **Multi-provider LLM enrichment** (GIT-110) — OpenAI, Gemini, Ollama alongside Anthropic; auto-detected from env vars or configured in `.git-mem.yaml`
+- **Multi-agent detection** (GIT-121) — auto-detects Claude Code and Codex from env vars, resolves model from config files (session JSONL, config.toml)
+- **`git mem trailers` command** (GIT-71) — inspect AI-* trailers on commits, query across history, list distinct keys
 
 ### Fixed
 
 - Edge cases in commit-msg hook (empty messages, missing env vars)
 - Duplicate Agent/Model trailers prevention
 - Hook chaining — all hooks now install by default
+- **Hook LLM config passthrough** (GIT-122) — hook command now forwards `llm` config from `.git-mem.yaml` to DI container
+- **Init MCP** — support source-checkout local server resolution (GIT-80)
+- **Init push semantics** — preserve default push semantics for notes refs (GIT-117)
+- **Circular error references** — in logger and handlers (GIT-109)
+- **Git root resolution** — for init paths + Windows test compat (GIT-108)
+- **Undefined content** — handle undefined content in memory query (GIT-96)
+- **Windows compat** — PowerShell-safe lint globs (GIT-118), cross-platform test commands (GIT-119)
 
 ## [0.3.0] - 2026-02-12
 

CLAUDE.md

@@ -37,7 +37,7 @@ Clean architecture with three layers. Dependencies point inward only: Infrastruc
 **Infrastructure** (`src/infrastructure/`) — Implements domain interfaces. `GitClient` wraps git CLI. `NotesService` reads/writes `refs/notes/mem`. `TrailerService` queries commit trailers. `MemoryRepository` persists `IMemoryEntity[]` as JSON in git notes. `HeuristicPatterns` provides regex-based extraction rules. `EventBus` provides pub/sub event dispatch with error isolation (failing handlers don't crash the hook).
 
 **Entry points:**
-- `src/cli.ts` — Commander.js CLI with 6 commands (remember, recall, context, extract, sync, init)
+- `src/cli.ts` — Commander.js CLI with 8 commands (remember, recall, context, extract, sync, init, trailers, hook)
 - `src/mcp-server.ts` — MCP server over stdio; `src/mcp/server.ts` creates the server and registers 4 tools
 - `src/commands/hook.ts` — Unified hook entry point: reads stdin JSON, loads config, emits typed event via EventBus
 - `src/commands/init.ts` — Interactive setup: hooks, MCP config, .gitignore, initial extract
@@ -51,7 +51,7 @@ const container = createContainer({ logger, scope: 'remember' });
 const { memoryService } = container.cradle;
 ```
 
-Uses `InjectionMode.CLASSIC` (matches constructor parameter names to registration names). `ICradle` in `types.ts` defines the typed container shape with all interface references. The container also registers the `EventBus` with three hook handlers (`session:start`, `session:stop`, `prompt:submit`) wired during creation.
+Uses `InjectionMode.CLASSIC` (matches constructor parameter names to registration names). `ICradle` in `types.ts` defines the typed container shape with all interface references. The container also registers the `EventBus` with five hook handlers (`session:start`, `session:stop`, `prompt:submit`, `commit:msg`, `post:commit`) wired during creation.
 
 **Hook event flow:** `git-mem hook <event>` → `readStdin()` → `loadHookConfig()` → `createContainer()` → `eventBus.emit(typedEvent)` → handlers return `IEventResult[]` → output to stdout (context), summary to stderr. Hooks have a 10s hard timeout and never throw — failures are caught and reported silently.
 

README.md

@@ -2,6 +2,8 @@
 
 Track AI contributions in your git history.
 
+![AI commit metadata](git-mem.png)
+
 ## Why?
 
 AI writes code now. Your commits should show when, which agent, and which model.

docs/getting-started.md

@@ -138,7 +138,7 @@ Options:
 - `--commit-count <n>` — Max commits to process
 - `--dry-run` — Preview without writing
 - `--threshold <n>` — Interest score threshold (default: 3)
-- `--enrich` — Enable LLM enrichment (requires `ANTHROPIC_API_KEY`)
+- `--enrich` — Enable LLM enrichment (requires an LLM provider API key)
 
 ### context
 
@@ -164,6 +164,24 @@ git mem sync --push   # push only
 git mem sync --pull   # pull only
 ```
 
+### trailers
+
+Inspect AI-* trailers on commits.
+
+```bash
+git mem trailers              # show trailers on HEAD
+git mem trailers abc123       # show trailers on specific commit
+git mem trailers --query AI-Decision  # search across history
+git mem trailers --keys       # list all distinct AI-* trailer keys
+```
+
+Options:
+- `--query <key>` — Search for trailer key across history (e.g. `AI-Decision`)
+- `--since <date>` — Filter commits after date
+- `--keys` — List all distinct AI-* trailer keys in the repo
+- `-n, --limit <n>` — Max commits to search
+- `--json` — Output as JSON
+
 ---
 
 ## Using Claude Code?
@@ -177,6 +195,8 @@ If you use Claude Code, git-mem can do more than just add trailers & notes — i
 | **SessionStart** | Loads stored memories into Claude's context on startup |
 | **Stop** | Extracts knowledge from commits made during the session |
 | **UserPromptSubmit** | Surfaces relevant memories per prompt (disabled by default) |
+| **PostCommit** | Records AI agent and model metadata on each commit |
+| **CommitMsg** | Analyzes commit message and adds AI-* trailers (with optional LLM enrichment) |
 
 These hooks read the AI trailers from your commits and build a searchable memory layer that Claude can query.
 
@@ -195,13 +215,35 @@ hooks:
   promptSubmit:
     enabled: false
     surfaceContext: true
+  commitMsg:
+    enabled: true
+    enrich: false          # set true + provide LLM API key for enriched extraction
+    enrichTimeout: 8000
+  postCommit:
+    enabled: true
 ```
 
 
 ## Environment Variables
 
 | Variable | Required | Description |
 |----------|----------|-------------|
-| `ANTHROPIC_API_KEY` | Only for `--enrich` | Anthropic API key for LLM enrichment during extract. Get one at [console.anthropic.com](https://console.anthropic.com/). |
+| `ANTHROPIC_API_KEY` | For Anthropic LLM | Anthropic (Claude) API key |
+| `OPENAI_API_KEY` | For OpenAI LLM | OpenAI API key (requires `npm install openai`) |
+| `GOOGLE_API_KEY` / `GEMINI_API_KEY` | For Gemini LLM | Google Gemini API key (requires `npm install @google/generative-ai`) |
+| `OLLAMA_HOST` | For Ollama | Ollama server URL (default: `http://localhost:11434`) |
+| `GIT_MEM_LLM_PROVIDER` | No | Force provider: `anthropic`, `openai`, `gemini`, `ollama` |
+
+Only one provider is needed. If `--enrich` is used without any LLM API key, git-mem falls back to heuristic extraction with a warning.
+
+### LLM Configuration
 
-If `--enrich` is used without an API key, git-mem falls back to heuristic extraction only.
+You can also configure the LLM provider in `.git-mem/.git-mem.yaml`:
+
+```yaml
+llm:
+  provider: openai       # auto-detected from env vars if omitted
+  model: gpt-4o          # uses provider default if omitted
+  intentModel: gpt-4o-mini
+  baseUrl: http://localhost:11434  # for ollama
+```

package.json

@@ -1,7 +1,7 @@
 {
   "name": "git-mem",
-  "version": "0.4.0",
-  "description": "Git-native memory layer for AI coding tools. Store decisions, context, and knowledge directly in your git repository using commit trailers and git notes.",
+  "version": "0.5.0",
+  "description": "Add AI metadata to git commits automatically. Tracks agent, model, and decisions via commit trailers and git notes.",
   "bin": {
     "git-mem": "dist/cli.js",
     "git-mem-mcp": "dist/mcp-server.js"

tests/integration/hooks/hook-commit-msg.test.ts

@@ -439,7 +439,10 @@ describe('Integration: hook commit-msg', () => {
     it('should detect Claude Code agent from CLAUDECODE env var', () => {
       delete process.env.GIT_MEM_AGENT;
       process.env.CLAUDECODE = '1';
-      process.env.ANTHROPIC_MODEL = 'claude-opus-4-5-20251101';
+      // Use GIT_MEM_MODEL (highest priority) rather than ANTHROPIC_MODEL,
+      // because detectClaudeModel() reads session JSONL files when CLAUDECODE
+      // is set, which can find real sessions in the dev environment.
+      process.env.GIT_MEM_MODEL = 'claude-opus-4-5-20251101';
 
       const commitMsg = 'feat: claude code commit';
       const msgPath = createCommitMsgFile(repoDir, commitMsg);
@@ -457,7 +460,7 @@ describe('Integration: hook commit-msg', () => {
         modifiedMsg.includes('Claude-Code') || modifiedMsg.includes('claude-code'),
         'Agent should be Claude Code variant',
       );
-      assert.ok(modifiedMsg.includes('AI-Model: claude-opus-4-5-20251101'), 'Should use ANTHROPIC_MODEL');
+      assert.ok(modifiedMsg.includes('AI-Model: claude-opus-4-5-20251101'), 'Should use GIT_MEM_MODEL');
     });
   });