| name | review-audit-bridge |
| description | Keep code review, security review, and QA passes evidence-based and stage-aware for this repo. Use when a ticket already has an approved plan and implementation evidence and a review lane needs structured findings. |
Review Audit Bridge
Before starting a code review, security review, or QA pass, call skill_ping with skill_id: "review-audit-bridge" and scope: "project".
Use this skill after implementation exists. It bridges the review and QA lanes so they return evidence-backed findings instead of vague commentary.
This is a generated repo-local skill. It may recommend remediation or reverification follow-up, but it does not become the canonical ticket owner by itself.
Prioritize findings in this order:
- correctness bugs
- behavior regressions
- security or trust issues
- missing tests or validation gaps
- maintainability concerns
Return:
- For code review or security review:
- Findings ordered by severity
- Risks
- Validation gaps
- Blockers or approval signal
- For QA:
- Checks run
- Pass or fail
- Blockers
- Closeout readiness
Rules:
- findings come first; do not open with praise or a summary
- reference exact files, diffs, commands, or artifact paths
- if the approved plan, implementation artifact, or required validation context is missing, return a blocker instead of inferring correctness
- use
ticket-execution for lifecycle order and project-context for canonical repo docs
- run the repo's build command, lint or type-check command, and reference-integrity checks when those quality gates exist for the active stack; include the raw command output in the review or QA artifact
- if any required build, lint, type-check, or reference-integrity command fails, the verdict must be FAIL and the artifact must name the failing command plus the concrete error
- for setup, bootstrap, or repo-foundation tickets, direct toolchain command output is required. Do not accept
JAVA_HOME or environment discovery as Java version proof; the artifact must include output from java -version or javac -version. Do not accept Android tool ls output as build-tool version proof; require aapt2 version, d8 --version, and zipalign usage/version output. If a command tool call lacks a completion/output record, treat it as not run, and reject permission-policy excuses, indirect proof through Godot, or validation deferred to another ticket.
- do not approve setup/bootstrap evidence that says required validation commands are "not yet applicable", "will run after" another ticket, or merely "identified" unless the current canonical ticket acceptance explicitly allows that deferral.
- when the active ticket owns a user-facing or runtime-backed surface, reject explicit TODO-only behavior, placeholder responses, shallow command shells, or stubbed integrations in the claimed code path even if compile/test checks pass
- when reviewing or validating a remediation ticket with
finding_source, rerun the original finding-producing command first, confirm the specific failure is gone, then check that adjacent quality gates still pass
- write any workflow-failure explanation or review retrospective to the repo-local process-log path described in
references/review-contract.md
- recommend follow-up tickets only when current evidence justifies them, and route canonical ticket creation through the repo's guarded ticket workflow
- do not claim that repo files changed
References
references/review-contract.md