Forge automation control
Skill jukrap/ai-agent-playbook/skills/ai-harness/forge-automation-control
Reusable AI agent skills, project templates, and guardrails for safer software maintenance and delivery.
npx -y skills add jukrap/ai-agent-playbook --skill forge-automation-controlAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when coordinating or automating resumable work through GitHub or Gitea issues, pull requests, actions, scheduled ticks, or local fallback runs.
SKILL.md
3.2 KB, as published. Nobody here has run it
Forge Automation Control
Use this skill to make forge-backed work resumable, reviewable, and safe when no forge is available.
Workflow
- Inspect repository state, remotes, authentication, configured provider, and the current deny flags before proposing any remote write.
- Resolve the effective permission profile. Apply the most restrictive user instruction, project setting, CLI flag, and provider capability.
- Validate an approved structured plan. Keep fine-grained execution tasks in the local ledger and publish reviewable delivery groups as issues; task-per-issue mode is an explicit legacy choice.
- Run one idempotent tick at a time: claim one ready task, execute it within budget, rerun verification in the controller, record evidence, and checkpoint state.
- Synchronize only meaningful transitions, blockers, reconciliation requests, and final verification. Reuse managed issues, marker comments, branches, and draft pull requests.
- Pause when requirements change during execution, a lease or permission is uncertain, verification fails repeatedly, or a high-risk action needs approval.
- Fall back to the local ledger when remote access is unavailable or denied. Never treat remote synchronization as a prerequisite for local progress.
Safety Boundaries
- Keep merge, release, delete, force-push, and protected-branch writes approval-bound in every profile.
- Keep forge credentials out of configuration, prompts, worker environments, ledgers, evidence, and logs.
- Do not let a worker push or mutate forge state; the controller reviews files and verification before delivery.
- Preserve the user's working checkout. Use a managed isolated checkout for unattended execution.
- Do not invent a merge-first, release-first, or other gate that is absent from the approved plan. A merge approval may hold merge while branch implementation, verification, and a draft pull request continue.
- Pause before the first write when a configured GitHub Project requires a missing
projectscope. Showgh auth refresh -s projectand the status recheck command; never run authentication refresh automatically.
Related Skills
- Use
issue-planning-triageto shape a roadmap into a small set of independently reviewable delivery-group issues and to classify pre-existing ready-label work. Keep finer execution tasks in the local ledger. - Use
agent-orchestration-handoffwhen executor or review workers need bounded contracts, evidence ownership, and reconciliation handoff. - Use
git-worklog-guardrailsbefore controller staging, commit, push, draft pull request, or durable worklog updates.
References
Read references/provider-capabilities.md when detecting GitHub or Gitea, planning remote artifacts, or choosing a capability fallback.
Read references/state-and-permissions.md when resolving task transitions, queue labels, progress, permission profiles, deny flags, or approval gates.
Read references/scheduler-and-recovery.md when configuring ticks, supervisors, Actions, local schedulers, leases, crash recovery, or unattended Git delivery.