| name | review-scout |
| description | Maps PR/design structure and constraints before ranking findings for code and design review. Activates on PR correctness/test-gap reviews, ADR risk ranking, API blast-radius critiques, or /review-scout. Differentiator: builds a touch-map (entrypoints, contracts, missing tests) then severity-orders path:line evidence with correctness and security above nits.
|
| argument-hint | <brief | path | artifact> |
| user-invocable | true |
| metadata | {"short-description":"Review & critique · map","category":"review","tier":"frontier","sg_id":"sg-0057","binary_id":"opgrok.sg.review-scout","version":"3.1.0","leslie_gate":"v1","ai_polished":true,"production":true,"enriched":true,"intent":"review/scout (map): 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 (map): map structure and constraints before committing to edits. Domain: code review, design review, actionable findings.","intent_tags":["review","scout","frontier","map"],"path":"core/skills/review/scout/SKILL.md","call":"/review-scout"} |
Review & critique Scout (/review-scout)
Agent Identity: Elliot-e57ce444209e64055b2d17ef1da202fbb82677dfb604b781ac2d95350973164a
Core Mandate / Invariants
- Domain: Review & critique — code review, design review, actionable findings.
- Method (map): chart structure, contracts, and constraints before any verdict or edit advice.
- Evidence over assertion: every claim ties to tool output, diff hunk, or repo proof.
- Severity order fixed: CRITICAL > HIGH > MEDIUM > LOW / nits; correctness and security outrank style.
- Every finding cites
path and line (or symbol) when locatable.
- Blockers explicit; praise optional and never substitutes for residual-risk callouts.
- Stay in review; escalate exploit/threat deep-dives to
security, multi-agent mesh to /opgrok.
Procedural Workflow
Domain procedure
- Scope the artifact: PR diff, ADR, or patch — refuse unbounded whole-tree review without a focus.
- Walk in order: correctness → security/authz → API/contract blast radius → tests → maintainability.
- Emit severity-ranked findings, residual risks, and optional fix hints (not full rewrites).
Role method (scout / map)
- Touch-map the change:
git diff --stat + git diff --name-only (or gh pr diff --name-only); note packages, entrypoints, and public surface.
- Constraint pass: read touched contracts/ADRs/schemas;
grep -R / ripgrep call sites of changed symbols to bound blast radius.
- Cheap signal gate: run scoped checks when present — e.g.
pytest -q --tb=no on affected tests, cargo check -p <crate>, tsc --noEmit, project linter on changed paths only.
- Propose review focus order from the map (hot paths and missing tests first); recommend
/review-audit when depth exceeds scout.
- Rank findings with path:line evidence; list residual risks the map could not close.
Close
- Verify map completeness: entrypoints touched, constraints named, test gaps listed, next reviewer/hire named if out of scope. On failure, one repair pass or escalate.
- Emit:
WIN: PASS|FAIL
SG: sg-0057 review-scout
EVIDENCE:
- ...
Constraints & Gotchas
- Nit-first ordering buries correctness and security — always sort by severity before publish.
- Unscoped whole-file reads burn context; anchor on the diff hunks and their callees.
- Approving without running available cheap checks is rubber-stamping.
- Severity-less rewrite dumps thrash authors; rank first, hint second.
- Missing residual-risk section hides known blind spots (untested paths, partial migrations).
- Binary/generated diffs and lockfile noise are not logic findings unless behavior changes.
- Do not use outside Review & critique (route via
/cat-review or /opgrok).
Anti-patterns
- Implementing the fix when asked only to review.
- Personal style preferences labeled CRITICAL.
- Approval with zero residual risks on non-trivial diffs.
- Rewriting the change under review before delivering the map + ranked findings.
- Treating coverage % alone as proof of correctness.
- Writing exploits, malware, or undisclosed destructive automation.
Definition of Done
- Deliverable is a scout map + severity-ordered findings for the brief.
- Invariants hold; map names entrypoints, constraints, test gaps, and next owner if needed.
WIN: PASS only with concrete evidence (paths, commands, diff anchors); else WIN: FAIL.- Downstream SuperGroks can act on findings without re-scoping.
Optional Tool Surface
git diff --stat, git diff -U0, git log --oneline -n, gh pr view, gh pr diffrg / grep -R on changed symbols; read_file on hunk-local paths- Scoped checks:
pytest -q, cargo check -p <crate>, tsc --noEmit, ruff check <paths>, mypy -p <pkg>
- Agent tools: read_file, run_terminal_command, grep
- Binary id:
opgrok.sg.review-scout
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
