git:20261025.83f68e3 to git:20261107.665a973

- Use at each required [execution checkpoint](../execute-plan/SKILL.md#completion-and-review-checkpoints), including phase checks before review and recovery after the selected review sequence, or for a standalone final check. Review the full checkpoint change set against the original request and later updates, including integration with completed phases. Resolve task-relevant omissions and inconsistencies across code, documentation, tests, and configuration within scope.

- For a pre-review phase checkpoint, complete phase acceptance, cleanup, and required checks, then return to execution with selected reviews pending. Pending review coverage does not prevent that phase checkpoint from closing. Post-review coverage recovery below applies only after the selected review sequence returns or the active loop ends at its cap or a concrete continuation blocker.

- Reconcile execution and worker residuals, review findings, failed checks, and verification gaps. Fix all authorized task-relevant issues, including pre-existing failures discovered during the task. A review cap or terminal read-only audit does not defer these repairs: perform them here without restarting the loop. Reject unsupported findings with evidence.

- Do not use this recovery to replace a fresh review still required and available within the active loop, including after round 3. A failed attempt within an active loop follows its [failure recovery policy](../code-review-loop/SKILL.md#recover-failed-attempts), including a required next fresh review after a verified correction. When the loop ends at its cap or a concrete continuation blocker, identify missing coverage and unreviewed repairs against the original target, acceptance criteria, and review lenses. Inspect that coverage directly and record the evidence in task context; do not retry the failed call, restart the loop, or add disguised review rounds. A checkpoint can close when this coverage and all required repairs and checks are complete. Preserve the original loop's incomplete status or unreviewed-fix limit in the report; this recovery does not establish a clean loop result. If required coverage cannot be established, keep the checkpoint incomplete and apply the stop rules below.

- Perform one cleanup round over changed and directly affected files. Remove confirmed obsolete, duplicate, dead, or unnecessary compatibility code and related tests/configuration/docs. Preserve required behavior, explicit compatibility, and unrelated work. With read-only or documentation-only authority, report material implementation findings without editing them. A clean result is valid.

- Discover the project's full testing and quality gates from repository instructions (including AGENTS.md), CI workflows, scripts, and contributor/testing documentation. Establish passing evidence for every required gate on the final relevant state during each invocation. Apply [Check evidence and reuse](references/check-evidence.md) to run a gate or reuse valid evidence from this task. Targeted tests, smoke tests, or document validators do not replace full suites or quality checks. Small changes and long runtime are not reasons to reduce coverage. Add checks proportional to risk where required gates leave a gap.

- Only explicit user instructions, documentation-only work, or checks that incur actual external charges permit skips. For documentation-only work, run required document or skill validators and skip unrelated application suites; this exception does not cover code, configuration, or test changes anywhere in the checkpoint. Skip paid checks, such as smoke tests that call billed APIs, unless already authorized; ordinary runtime and local compute are not paid-check exceptions. Establish passing evidence for all remaining free coverage under the shared reuse policy. Preserve authorization boundaries for external effects; missing access, services, or authority blocks a required check rather than exempting it. Try reasonable authorized alternatives and report unresolved blockers.

- Fix authorized in-scope failures and verify repairs. Reassess check evidence after every relevant change and rerun invalidated gates under the shared reuse policy. A failed or blocked required gate keeps the checkpoint incomplete. Record each required gate's command, result (passed, failed, skipped, or blocked), whether passing evidence was executed or reused and its origin, and the reason for any skip or blocker.

- After repairs, inspect the affected result and verify it again as needed until no actionable task-relevant issue remains and required checks pass. A phase checkpoint does not end execution: return to the executor to continue the remaining plan. Stop with work remaining only for user direction, an explicit authorized session boundary, or a concrete blocker requiring missing authority, input, or external change; first try reasonable authorized alternatives and complete independent work. Report blocked work as incomplete, not a completed result with residuals.

- Lead with changes or findings, then report executed and reused check results, explicit skips, and blockers. Use short, plain prose without stock headings or repeated summaries. Do not claim implementation verification from document validators or fresh review from reusable check evidence. Save a user report only if requested; check logs follow the temporary-file rules.

- Keep scratch in context unless files serve a concrete agent need; never create them solely for user presentation. Track task-owned paths; remove them after use, including failure or abandonment, and verify cleanup before returning. Retain files only for active agent continuation with a cleanup owner and removal point; report retained paths or cleanup failures. Preserve requested deliverables (including `.local/` plans), pre-existing files, and unrelated work; never delete shared directories wholesale.