Skip to content

Repository files navigation

🐾 Ferret

Claude Code plugin for local-first time tracking in AI-assisted development. Parallel task analysis · Session lifecycle · Token metrics · Git-native

Install

Jednorazowa sesja (najszybciej)

claude --plugin-dir /absolutna/sciezka/do/ferret

Permanentna (wszystkie projekty)

# Temporary (dev/test):
claude --plugin-dir ./agent-tracker

# Permanent (your user profile):
claude plugin install ./agent-tracker --scope user

# Per-project (committed to repo):
claude plugin install ./agent-tracker --scope project

Then use /ferret: commands in any Claude Code session.

Commands

Command What it does
/ferret:plan Analyse task dependencies → find what can run in parallel
/ferret:start PROJ-123 Start tracking a task
/ferret:pause / /ferret:resume Context switch
/ferret:done Complete session
/ferret:status Show active sessions
/ferret:annotate Append Claude-Time: trailer to last commit
/ferret:brag (optional) Generate XYZ eval / case-study / blog
/ferret:score Parallelism · velocity · token efficiency report

/ferret:plan is read-only

It only analyses and reports. It does not create worktrees, branches, or sessions. Orchestration is yours – use whatever SDLC tooling your team uses.

/ferret:plan --source linear

📋 PARALLELIZATION ANALYSIS
✅ CAN RUN IN PARALLEL: PROJ-123, PROJ-127, PROJ-131
⛔ MUST BE SEQUENTIAL: PROJ-124 [needs: PROJ-123]
Critical path: PROJ-123 → PROJ-124 (5h minimum)
Speedup opportunity: 2× vs sequential

Tracking sessions

Sessions are plain JSON files in data/sessions/. Nothing leaves your machine.

/ferret:start PROJ-123 --estimate 2h
# ... work ...
/ferret:annotate          # add Claude-Time: to last commit
/ferret:done              # complete, write checkpoint

Git integration

feat: implement auth service

Claude-Time: elapsed=2h15m tokens=45231 session=abc12345 concurrent=2

Optional hook for automatic trailer on every commit:

cp hooks/prepare-commit-msg /your-project/.git/hooks/
chmod +x /your-project/.git/hooks/prepare-commit-msg
export FERRET_DIR=/path/to/agent-tracker

Brag docs (optional)

/ferret:brag --style case-study --period "last-30-days"
/ferret:brag --style blog --tasks PROJ-123

Uses Google XYZ Framework: "Accomplished [X] as measured by [Y], by doing [Z]." Extended for AI: token cost and concurrency as Y metrics.

Scoring

overall = 0.35 × parallelism + 0.30 × velocity + 0.25 × token_efficiency + 0.10 × quality
/ferret:score --compare   # current vs previous period

Open source · Local-first · No external services required Works with any LLM harness, not just Claude Code

About

Claude Code plugin for local-first time tracking in AI-assisted development. Parallel task analysis · Session lifecycle · Token metrics · Git-native

Resources

Stars

Watchers

Forks

Releases

Sponsor this project

Contributors

Languages