Skip to content
worca

How it works

A worca run moves a task through a sequence of specialized agents, each with one job. They hand off down the pipeline — and loop back when work needs another pass — until a pull request exists.

The pipeline

Section titled “The pipeline”
critical issuestests failchanges requested Preflight Validates the environment Planner Explores code, writes the plan Plan Reviewer Audits the plan (optional) Coordinator Splits the plan into tasks (aka beads) Implementer Builds each task, test-first Tester Runs tests, collects proof Reviewer Reviews code and conventions Guardian Commits, opens the PR Learner Retrospective (optional)

Hover or focus any stage to see what it does. Plan Review and Learn ship disabled — enable them per project when you want the extra rigor.

Governance at every step

Section titled “Governance at every step”

worca enforces its rules through hooks that run on every tool call:

  • only the Guardian may run git commit;
  • dangerous operations (rm -rf, force-push, environment writes) are blocked;
  • a test gate halts the run after repeated test failures;
  • each agent can only dispatch the tools, skills, and subagents it’s allowed.

A live dashboard

Section titled “A live dashboard”

The worca-ui dashboard streams the whole run over WebSocket — stages, per-iteration cost, token usage, and logs, with pause / resume / stop controls. Nothing is polled; the page updates as the pipeline moves.