From issue queue
to reviewed PR.
AI pipeline that plans, reviews, executes, tests, verifies, and opens the PR from a GitHub issue.
Install in seconds
Homebrew for the desktop app. npx for the CLI. Same pipeline engine underneath.
Four checkpoints. One pipeline.
Every issue goes through planning, review, execution, testing, verification, and shipping before it becomes a PR.
Plan
Drop a GitHub issue. The planner reads your codebase, generates a task graph, and proposes an execution strategy.
Supports Claude, Codex, and OpenRouter models per phase.
Review
A separate reviewer LLM critiques the plan — checking for missed edge cases, security issues, and scope creep.
Multi-round revision until the plan is stable.
Execute
Each task runs in an isolated git worktree. The agent writes code, runs tests, and self-corrects on failures.
Worktree isolation keeps local branches and dirty state separated.
Verify + Ship
Test output, runtime QA, visual QA evidence, and full diff review are checked before ShipCode opens the PR.
Automatic retries and PR stabilization handle blockers before handoff.
What you get out of the box
A full-stack pipeline engine, not another chat wrapper.
Multi-model routing
Claude, Codex, or OpenRouter: pick the best model for each pipeline phase. Swap providers without rewriting prompts.
Automations and scheduling
Run cron-like recurring prompts, auto-run Todo work, and drain queued jobs through workspace and project capacity caps.
Task graph execution
Complex issues can decompose into a dependency graph. Tasks execute in topological order with per-node context.
Isolated worktrees
Every pipeline runs in its own git worktree. No branch conflicts, no stash juggling, no dirty state leaking between tasks.
Runtime QA with retries
Typecheck, test suite, browser/runtime QA, and visual QA evidence feed the verifier before a PR ships.
Desktop control surface
Track issues, PRs, worktrees, code, costs, insights, terminal output, and scheduled automations in one app.
Ship with agents, not arguments.
Open-source. Runs locally. Your code never leaves your machine.