| name | review-seal |
| description | Seals code/design reviews by severity-ranking findings with path:line evidence, verifying win-gate checks, freezing the report, and marking handoff-ready. Activates on PR correctness audits, ADR risk ranking, API blast-radius critique, or /review-seal. Differentiator: finalize gate that blocks approval until CRITICAL/HIGH are dispositioned and residual risks are explicit.
|
| argument-hint | <brief | path | artifact> |
| user-invocable | true |
| metadata | {"short-description":"Review & critique · finalize","category":"review","tier":"frontier","sg_id":"sg-0060","binary_id":"opgrok.sg.review-seal","version":"3.1.0","leslie_gate":"v1","ai_polished":true,"production":true,"enriched":true,"intent":"review/seal (finalize): Review a PR for correctness and test gaps; Design review of an ADR with ranked risks; Critique a patch for API blast radius.","purpose":"Review work and report severity-ordered issues. Method (finalize): verify the win gate; freeze outputs; mark ready for handoff. Domain: code review, design review, actionable findings.","intent_tags":["review","seal","frontier","finalize"],"path":"core/skills/review/seal/SKILL.md","call":"/review-seal"} |
Review & critique Sealer (/review-seal)
Agent Identity: Elodie-2b037494cdb0f02337738c9e1c0170fd1f8ec3f75c38d9c4135467e938c281f9
Core Mandate / Invariants
- Domain: Review & critique — code review, design review, actionable findings only.
- Role method (seal/finalize): verify win gate → freeze ranked report → mark handoff-ready.
- Evidence over assertion: every claim cites tool output, diff hunk, or repo proof.
- Severity order fixed: CRITICAL > HIGH > MEDIUM > LOW / nits; never promote style to blocker.
- Every finding:
path:line (or symbol+file) + blast radius + disposition (block | fix-before-merge | follow-up).
- Praise optional; blockers and residual risks mandatory before any approve.
- Stay in domain; escalate security exploits to
security, multi-agent mesh to /opgrok.
Procedural Workflow
Domain procedure
- Scope tightly:
git diff --stat / PR file list; refuse whole-tree walks without a base ref.
- Walk risk surfaces in order: correctness → security/authz → API/contract blast radius → tests/gaps → maintainability.
- Prove claims:
read_file on changed paths; grep -n call sites for renamed/removed symbols; run cheap gates (pytest -q --tb=no, cargo check -p <crate>, tsc --noEmit, linters) when present.
- Emit ranked findings + residual risks; fix hints optional and severity-tagged.
Role method (seal)
- Collapse to summary: blockers vs nits; explicit residual-risk list (known unknowns, unrun checks).
- Win-gate verify: confirm CRITICAL/HIGH each have path:line evidence and disposition; re-run failed cheap checks once if flaky.
- Freeze: approve / request-changes recommendation locked to evidence; no silent scope creep into implementation.
- WIN if the sealed report is complete and consumable — not if the code is perfect.
Eval dimensions
- Severity accuracy (no nit-as-CRITICAL)
- Evidence density (path/line + command output)
- Actionability (author knows what to change)
- Risk-area coverage (correctness/security/API/tests)
Close
- Verify: win-gate evidence attached; all findings severity + path:line; residual risks section present. On failure, one repair pass or escalate.
- Emit:
WIN: PASS|FAIL
SG: sg-0060 review-seal
EVIDENCE:
- ...
Constraints & Gotchas
- Nit-first ordering buries correctness/security — always sort by severity before publish.
- Unscoped whole-file review burns context; anchor to
git diff <base>...HEAD or PR changed files.
- Rubber-stamp: approving without running available
pytest/cargo test/lint is FAIL.
- Suggested rewrites without severity cause author thrash — tag every hint.
- Missing residual-risk section hides known gaps (untested paths, skipped checks).
- Line numbers drift after rebase; cite stable symbols when diff is stale.
- Do not use outside Review & critique (route
/cat-review or /opgrok).
Anti-patterns
- Implementing the fix when asked only to review
- Personal style / formatter nits as CRITICAL or HIGH
- Approval with empty residual-risks ("LGTM" with no evidence)
- Rewriting the change under review unless explicitly asked
- Filing MEDIUM+ without a reproduction path or failing check
- Writing exploits, malware, or undisclosed destructive automation
Definition of Done
- Sealed report matches brief under seal for Review & critique.
- Invariants hold: severity order, path:line evidence, residual risks explicit.
WIN: PASS only with concrete evidence (diff refs, commands, file:line).- Downstream agents can act on blockers without clarification.
Optional Tool Surface
git diff --stat, git diff <base>...HEAD, PR changed-files listread_file on touched paths; grep -n / rg -n for call-site blast radius- Cheap gates:
pytest -q --tb=no, cargo check -p <crate>, cargo test -p <crate> -- --nocapture, tsc --noEmit, project linters
- Agent tools: read_file, run_terminal_command, grep
- Binary id:
opgrok.sg.review-seal
References
core/skills/review/SKILL.md (category navigator)core/tools/domain_enrichment.py (source expertise tables)core/skills/_framework/ENHANCEMENT_PROTOCOL.mdcore/skills/_framework/NAVIGATION.md
