A cc-wf-studio development-loop guide for one autonomous iteration: guard existing work, handle urgent maintenance, propose and judge a user-visible improvement, then build and record it through an auto-dev pull request.
Tasks
Check for an existing open agent PR targeting auto-dev and steward it before starting another task.
Handle red integration CI or ci-failure issues, actionable security findings and human-reported bug issues in the documented priority order.
Orient using IMPLEMENTATION_PLAN.md, docs/progress-log.md, idea issues and unfinished local work; invent 3–5 proposals, judge them and select one.
Implement one qualifying improvement on a branch from auto-dev, record its outcome before committing and squash-merge only after CI is green.
Inputs
The cc-wf-studio repository, fresh origin/main and origin/auto-dev, current git status and any unfinished local work.
IMPLEMENTATION_PLAN.md containing the North Star, value axes and not-value list; docs/task-automation.md and docs/progress-log.md.
Open PRs and CI results, idea/ci-failure/bug issues, actionable vulnerability findings, and the repository’s CLAUDE.md and PR conventions.
Outputs
For a passing proposal: one reviewable change and progress-log entry in the same commit, a changeset when required, and a PR targeting auto-dev.
For a non-passing iteration: up to three documented idea issues and a progress-log outcome, with no filler implementation.
For an unresolved PR: a re-armed CI check-in or an open needs-attention PR with a blocked log outcome after the retry limit.
Limitations and checks
Execution is serialized with capacity one, and one invocation addresses one task or existing PR.
No pushes to main, no PRs based on main and no release actions. Promotion from auto-dev to main is reserved for a human.
All five value-bar conditions must hold: value-axis alignment, a user-noticeable result, one-iteration scope, reversibility without breaking API/schema or release work, and a code-verified premise.
The skill must not change the North Star/value axes, repeat done or abandoned work, force a failing merge, or invent a task when none qualifies. Sync conflicts, missing credentials or ambiguous requirements require a human decision and a logged blockage.
Is an existing auto-dev PR being handled before another task begins?
Does the selected proposal meet every documented value-bar condition and have a verified code premise?
Was the progress-log entry included before the commit and PR, with build then check and the appropriate changeset?
Is the PR based on auto-dev, with CI completed and green before squash merge?
AI Search
Find projects, verify facts, compare options, or turn a complex need into an actionable plan
Try a searchA click only fills the search box; you stay in control