Worklane — Git skill for Claude Code
Open-source harness for running a crew of Claude agents on a real codebase: parallel workers, an independent evaluator, a shared GitHub Issues backlog, and a live dashboard showing who owns every task.
How to install Worklane
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open jayden1711/worklane and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Worklane does
Open-source harness for running a crew of Claude agents on a real codebase: parallel workers, an independent evaluator, a shared GitHub Issues backlog, and a live dashboard showing who owns every task.
Alternatives in Git
- Codebase To Course — by Zara Zhang - A Claude Code skill that turns any codebase into a beautiful, interactive single-page HTML cou 2.1k ★
- My Claude Code Setup — GitHub Repo stars Shared starter template configuration and CLAUDE.md memory bank system Setup template 2.1k ★
- Pull PRs — Pull open GitHub PRs into the github-pr tracker as backlog items 1.6k ★
README
Worklane
**Know who's on every task.**
Open-source harness for running a crew of Claude agents on a real codebase: parallel workers, an independent evaluator, a shared GitHub Issues backlog, and a live dashboard showing who owns every task.
**Status: pre-alpha (step 1 of 7).** Config, `install`/`doctor`, guardrail hooks, secret scanning, the vacuity check and the Stop gate work. The coordinator arrives in step 2.
What it is
- Parallel workers, each in its own git worktree, picking up
readyissues from a GitHub Issues backlog that humans and agents share. - A deterministic coordinator (plain code, not an LLM) that claims work, enforces budgets, retries, and moves state. Agents never hold GitHub write tokens; they propose actions and the coordinator validates and performs them.
- An independent evaluator with fresh context and read-only access that writes a failing reproduction test first, then judges the change against the issue's
done_whencontract. - Risk-based review levels computed from what a change touches, so docs fixes land on their own and money-path changes wait for a human.
- A land queue (direct-to-main or PRs) that batches compatible changes and bisects on failure.
- Guardrails that block the few truly dangerous things (signing from a live wallet, writing to the production database) and nothing else.
- One append-only event log that the dashboard, reports, scorecard and decision inbox all read from.
See [docs/design.md](docs/design.md) for the architecture and [docs/prior-art.md](docs/prior-art.md) for what we borrowed and from where.
Quick start (planned)
npm install --save-dev github:jayden1711/worklane#v0.1.0
npx worklane install # scaffolds .worklane/, hooks and settings
npx worklane doctor # verifies the install
npx worklane up # starts the coordinator and dashboard
Secret scanning
- CI and agent sessions: gitleaks scans every commit (agents can't commit unscanned) and
Related Skills
Epic Plan
Plan a new epic against a real codebase and a real project board, then — only after explicit approval — create
Agenteye
A real-time dashboard for all your GitHub Copilot CLI & Claude Code sessions: live state, checkpoints, history
Fleet Collect
Merge a fleet run's findings into one ranked backlog. Use after a run's workers have finished, or when asked w
Squad Fleet
Run multiple independent happysquad-loop tasks in parallel, each in its own git worktree and branch. Useful fo
Droid Office
A 3D office in your browser where your team hires Factory Droid, Claude Code, Codex and OpenCode workers at de
Dsh Skills Anywhere
Your skills, anywhere. Live Agent Skills provider for DeepSeek Harness (dsh): every agent's skill dirs (Claude
Related Agents
Gan Evaluator
GAN Harness — Evaluator agent. Tests the live running application via Playwright, scores against rubric, and p
Status Reporter
Use this agent to generate a cross-project status dashboard showing all projects at a glance. Run weekly or on
Logo Critic
Blind, independent scorer for logo candidates, especially AI-generated ones. Use after any generation round or