Awesome Reviewers expert instructions

domains / / openai/codex-security

Validate persisted selections

When accepting previously recorded state (digests, coverage manifests, etc.), revalidate it against the current computed/authoritative data before proceeding. If the value no longer matches, fail fast with a message that tells the user exactly what to do next. If you tighten validation rules, preserve backward compatibility by versioning the stricter...

raw .md Error Handling Python

When accepting previously recorded state (digests, coverage manifests, etc.), revalidate it against the current computed/authoritative data before proceeding. If the value no longer matches, fail fast with a message that tells the user exactly what to do next. If you tighten validation rules, preserve backward compatibility by versioning the stricter behavior (or adding legacy/empty allowances) so older artifacts aren’t rendered unusable.

Example (digest revalidation with clear failure):

def require_committed_diff_digest(target, base, head, content_digest=None):
    current = committed_diff_content_digest(target, base, head)
    if content_digest and content_digest != current:
        raise SystemExit(
            "The committed changes selected for review no longer produce the same "
            "diff. Select the changes to review again."
        )
    return current

Example (compatibility-aware validation):

  • If a future rule rejects empty/all-inapplicable coverage, gate it behind a version or explicit “authoritative empty inventory” signal; otherwise keep legacy scans readable so valid historical outputs don’t fail during recovery or re-ingestion.