PR

Own the decision report and authorized PR preparation/publication, not implementation, independent review, acceptance, merge, issue closure, deployment, release or installed-skill replacement. Loading this skill grants no effects. A report-only request stops at the report.

1. Establish the reporting contract

Identify target, audience, requested report or exact PR action, approved behavior/exclusions, agreement revision and acceptance IDs, comparison baseline, branch/base/PR target, protected work, evidence location and permitted disclosure. Read target instructions and relevant workflow, documentation, tracker-completion and release policies. Ordinary requests supply these facts through conversation and inspection; no handoff form is required.

For a bounded handoff, consume the caller's agreement, deciding sources/revisions (including dirty files), checks/environment, confirmed decisions/gaps, effect authority, ownership/state and expected return. Inspect the actual artifacts and raw results, not just the caller's preferred verdict. Preserve existing build/accept/report gate decisions and finding lineage; this package does not decide them anew.

Distinguish authority to draft, retain a local report, commit/push a branch, and create/update a remote PR. Use only effects approved by the task and project policy; an approved implementation may include PR publication if that policy expressly permits it. Unknown authority leaves the report local/conversational and publication blocked. Publication also needs permission for the exact content disclosed.

Complete when: scope, baseline, evidence sources, protected work and requested effects are known, or their specific blockers are recorded.

2. Inspect delivered evidence and its freshness

Inspect the full relevant diff/files, including committed, staged, unstaged and included untracked changes, renames, deletions, modes and symlinks. Identify the report's state separately from the remote PR's head; local dirty work is not remotely delivered. Record revisions/content digests, supporting inputs and environment. Protect unrelated work rather than staging or cleaning it.

Read current criterion/risk evidence and separate Standards/Spec reports, their identities, scope, findings and limitations when supplied. Reuse current complete evidence without restarting analysis or review. Compare deciding agreement/code/config/test/environment identities and coverage. Changed inputs invalidate dependent claims; retain old results for their old state. Unknown impact makes all relevant evidence stale. A matching HEAD alone does not establish dirty-work freshness.