What problem does it solve? Maintainers of the Senzing Bootcamp Claude Plugin need a safe way to keep development moving while they are away, without an autonomous agent making judgment calls, filing issues, or shipping unreviewed decisions in their name. ## Core Features & Use Cases - Label-gated autonomy: Works only GitHub issues carrying the unattended-ok label, treating an unlabeled backlog as a deliberate no-op rather than guessing what is safe. - Alternating implement/audit loop: Drives /implement-github-issue and /production-readiness-audit in up to five cycles, stopping when an audit records no new findings. - Explicit decision contract: Encodes what an unattended run may decide (issue order, implementation approach, reverting its own work) and what it must never decide (signing off invariants, filing issues, calling submit_feedback, declining work). - Cold-readable handoff: Ends every run with a written summary of commits, deferred invariants with drafted wording, blocked issues, audit verdicts, and self-reported mistakes. - Use Case: A maintainer labels three safe issues unattended-ok, confirms push policy before leaving, and returns to find each issue implemented and committed, an audit ledger entry per cycle, and a handoff listing exactly what needs their approval. ## Quick Start Run the unattended-issue-loop skill to work through my GitHub issues labeled unattended-ok while I am away and leave me a handoff summary.
