Automation Readiness Reference

July 27, 2026 ยท View on GitHub

Use this after loop-discovery.md selects automation, schedule-ready, or a scheduled/background loop. This file does not create automations; it defines the minimum contract before a report or skill recommends one.

Readiness Gate

Recommend automation only when all fields are concrete:

  1. Target: repository, path, issue, PR, report artifact, or external object.
  2. Trigger: cadence, cron-like schedule, event, or one-time follow-up.
  3. Run scope: standalone fresh run or thread-attached continuation.
  4. Execution location: local checkout, background worktree, CI, or external host; name why this location is safe enough.
  5. Input pack: exact command, prompt, report row, finding id, artifact, query, or MCP resource the run starts from.
  6. Sandbox: read-only, restricted write, full access, or explicit human approval for side effects.
  7. Validation: command, report quality check, CI result, review result, trace/eval comparison, or needs more evidence outcome.
  8. Triage path: where findings land and who reviews them.
  9. Risk boundary: files, systems, credentials, external writes, and destructive actions the automation must not touch without approval.
  10. Stop condition: state, count, date, score, passing validation, repeated covered result, or human decision that ends the loop.

If any field is missing, return needs more evidence or recommend a one-time manual follow-up instead of recurring automation.

Automation Shapes

  • Standalone automation: use when each run should start fresh from stable inputs and report findings into triage.
  • Thread automation: use when the scheduled work must preserve the current conversation or approval context.
  • Background worktree: use when the run may write files and must avoid unfinished local edits.
  • Local checkout: use only when direct access to the current workspace is required and the write risk is acceptable.
  • CI/GitHub Action: use when the loop is tied to repository events, protected checks, or branch policy.

Output Addendum

For schedule-ready outcomes, fill these slots:

Automation readiness: target=...; trigger=...; run scope=...; location=...;
input=...; sandbox=...; validation=...; triage=...; risk=...; stop=...

Do not call generic improvement advice automation-ready.