git-memory 0.1.0

package/skills/git-memory-index/SKILL.md

@@ -0,0 +1,88 @@
+---
+name: git-memory-index
+description: "Use when the user wants to index a repository into git-memory, start using git-memory on a new project, or re-index after significant history. Examples: \"index the lokumcu repo\", \"set up git memory for this project\", \"re-index everything\""
+---
+
+# Index a Repository with git-memory
+
+## When to Use
+
+- Starting git-memory on a new repository for the first time
+- After a major history rewrite or large batch of commits
+- User explicitly asks to (re-)index a project
+- Setting up git-memory for a team member's repo
+
+## Workflow
+
+```
+1. Confirm the repo path with the user (default: current directory)
+2. Run a dry-run first to show what will be indexed
+3. Ask if the user wants to limit commit count (useful for very large repos)
+4. Run the real index
+5. Run git-memory install to register the MCP server
+```
+
+## Checklist
+
+```
+- [ ] Confirmed repo path
+- [ ] Ran dry-run to preview commit count and filter results
+- [ ] Agreed on --limit if repo has >1000 commits
+- [ ] Completed indexing with 0 errors
+- [ ] MCP server registered in .claude.json
+- [ ] Reminded user to restart Claude Code
+```
+
+## Commands
+
+### Dry run (preview what will be indexed)
+```bash
+git-memory index \
+  --repo-path /path/to/repo \
+  --user-id my-repo-name \
+  --dry-run
+```
+
+### Full index
+```bash
+git-memory index \
+  --repo-path /path/to/repo \
+  --user-id my-repo-name
+# Add --limit 500 for large repos (indexes 500 newest commits)
+```
+
+### Install Claude Code plugin
+```bash
+git-memory install \
+  --repo-path /path/to/repo \
+  --user-id my-repo-name
+```
+
+### Or from within the repo directory
+```bash
+cd /path/to/repo
+git-memory index --user-id my-repo-name
+git-memory install --user-id my-repo-name
+```
+
+## Expected Output
+
+```
+Found 42 commits to evaluate
+Done. {"total_evaluated":42,"stored":39,"skipped_irrelevant":3,"skipped_duplicate":0,"errors":0}
+
+── Indexing Summary ─────────────────────────────
+  total_evaluated           42
+  stored                    39
+  skipped_irrelevant        3
+  skipped_duplicate         0
+  errors                    0
+```
+
+## Notes
+
+- **Skipped irrelevant**: Commits with no signal keywords (e.g. "k", "wip", "merge") are skipped by design. This keeps the index signal-rich.
+- **Deduplication**: Safe to run multiple times — ChromaDB deduplicates by commit hash.
+- **Large repos**: For repos with >10k commits, use `--limit 1000` to index the most recent 1000.
+- **user-id**: Use a distinct ID per repo so multiple repos can coexist in the same index.
+- **Embed providers**: Defaults to Ollama (nomic-embed-text). Set GIT_MEMORY_EMBED_PROVIDER=openai for OpenAI or =fastembed for fully local (no Ollama needed).

package/skills/git-memory-search/SKILL.md

@@ -0,0 +1,92 @@
+---
+name: git-memory-search
+description: "Use when the user asks about why code was written a certain way, wants to find commits related to a bug, feature or module, or needs historical context before editing. Examples: \"why does the payment module use a state machine?\", \"what commits touched auth?\", \"find all discount-related bug fixes\""
+---
+
+# Search Git History with git-memory
+
+## When to Use
+
+- User asks *why* something was built a certain way
+- User is about to edit a module and needs context
+- User reports a bug that may have been fixed before
+- User asks about a past decision, refactor, or migration
+- You need to understand the evolution of a component
+
+## Workflow
+
+```
+1. Identify the topic/component from the user's question
+2. Call search_git_history(query, limit=8)
+3. If results are few, also call bug_fix_history(component)
+4. If the user mentions a specific file, call commits_touching_file(filename)
+5. Synthesise results into a clear narrative for the user
+```
+
+## Checklist
+
+```
+- [ ] Called search_git_history with a descriptive query
+- [ ] Applied category filter if user is looking for a specific type (fix, feat, security)
+- [ ] Checked commits_touching_file if a specific file is involved
+- [ ] Included commit hashes in the answer so user can run git show
+- [ ] Noted the date range of relevant commits
+```
+
+## Tool Details
+
+### search_git_history
+```
+search_git_history(
+    query="payment state machine race condition",
+    limit=8,
+    category="fix"   // optional: fix | feat | security | refactor | migration | arch | perf
+)
+```
+Returns commits ranked by cosine similarity (0–1 score). Score > 0.7 = highly relevant.
+
+### commits_touching_file
+```
+commits_touching_file(
+    filename="PaymentService.ts",   // partial match supported
+    limit=10
+)
+```
+Returns all commits that modified the file, newest first, enriched with ChromaDB category.
+
+### bug_fix_history
+```
+bug_fix_history(
+    component="payments",
+    limit=8,
+    include_security=true
+)
+```
+Returns fix/bug/hotfix/security commits for the component, ranked by relevance.
+
+### architecture_decisions
+```
+architecture_decisions(
+    topic="state machine order transitions",
+    limit=5
+)
+```
+Returns refactor/migration/arch commits. Useful for "why is this designed this way?" questions.
+
+## Example
+
+**User:** "Why does the payment flow go through a state machine?"
+
+```
+1. search_git_history("payment state machine order transitions", limit=5)
+   → [32c3389f] fix(payments): route 3DS callback through transitionStatus
+     score=0.633 — "Replace direct repository updates with transitionStatus
+     calls to enforce the allowed-transitions guard"
+
+2. commits_touching_file("PaymentService.ts", limit=5)
+   → [32c3389f] fix(payments): route 3DS callback...
+   → [599d531b] feat: add Garanti 3DS payments...
+
+3. Answer: "The state machine was introduced in commit 32c3389f to prevent
+   race conditions on 3DS callbacks."
+```

package/skills/git-memory-status/SKILL.md

@@ -0,0 +1,69 @@
+---
+name: git-memory-status
+description: "Use when the user wants to check what is indexed, how many commits are in memory, or whether git-memory is set up correctly. Examples: \"is git memory set up?\", \"how many commits are indexed?\", \"check git memory status\""
+---
+
+# Check git-memory Status
+
+## When to Use
+
+- User asks if git-memory is configured
+- Before using search tools to confirm the index has data
+- Troubleshooting empty search results
+- After indexing to confirm it completed correctly
+
+## Workflow
+
+```
+1. Run status command to check ChromaDB commit count
+2. Check if MCP server is configured in .claude.json
+3. Check embed provider availability
+4. Report findings and any gaps
+```
+
+## Checklist
+
+```
+- [ ] ChromaDB has > 0 commits indexed
+- [ ] git-memory MCP server is in .claude.json
+- [ ] user-id matches between index and MCP server config
+- [ ] Embed provider is available (Ollama running / OPENAI_API_KEY set)
+```
+
+## Commands
+
+### Check index size
+```bash
+git-memory status --repo-path /path/to/repo
+```
+
+### Check MCP config
+```bash
+cat .claude.json | python3 -m json.tool | grep -A8 "git-memory"
+```
+
+### Check Ollama is running (required for default mode)
+```bash
+curl -s http://localhost:11434/api/tags
+```
+
+## Healthy Status Example
+
+```
+── git-memory status ──────────────────────────
+  Repo          : myproject
+  Chroma docs   : 39 (this repo) / 39 (all repos)
+  Total commits : 42
+  Coverage      : 93%
+  Embed provider: ollama
+```
+
+## Common Issues
+
+| Symptom | Fix |
+|---------|-----|
+| `0 commits indexed` | Run `git-memory index --repo-path .` |
+| MCP tools return errors | Check Ollama is running: `ollama serve` |
+| Wrong user-id | Must match `GIT_MEMORY_USER_ID` in MCP config |
+| Chroma path conflict | Ensure `GIT_MEMORY_CHROMA_DIR` points to `chroma_commits/` |
+| No LLM context (Layer 2) | Normal without Ollama — Layer 1 (ChromaDB) still works |