| name | phenomenal-review |
| description | Adversarial correctness/contract review of a diff using docs/agent/phenomenal-review.md (split-context stance; optional mode: multi for a second independent pass). Writes findings and follow-up tasks into docs. Use when the user asks for a phenomenal review, deep review, adversarial review, contract review, correctness review of local/branch/PR changes, or runs /phenomenal-review. Not the pre-ship fix loop and not a maintainability/code-judo pass.
|
| metadata | {"short-description":"Adversarial correctness review → docs follow-ups"} |
Phenomenal review (Grok wrapper)
This skill is a thin adapter. The full protocol is tool-agnostic:
docs/agent/phenomenal-review.md
Instructions
- Read
docs/agent/phenomenal-review.md in full (do not improvise a lighter bar).
- Execute that protocol against the user’s target:
- Default: uncommitted local changes (staged + unstaged + untracked sources).
- Or: branch, PR, or path list if the user named one.
- Mode:
standard (default) or multi if the user asks for multi/2×/adversarial pair.
- Adversarial stance — assume the code is wrong; you are reviewer, not implementer. Prefer diff-first. If this session also implemented the change, prefer
mode: multi (fresh subagent using the Pass B template in §3.7.1 of the protocol).
- Enforce sibling-path, reachability→severity, and primary evidence rules from the protocol (do not chain-trust prior review quotes or approximate counts).
- Review only — do not modify production or test code unless the user explicitly asks to harden afterward.
- Deliverables (both required by the protocol):
- Structured review note under
docs/ (suite-local or docs/agent/reviews/).
- Checkable follow-up tasks under
docs/ (not chat-only), including process fixes when bug classes recur.
- Report verdict, issue counts, file paths, and top issues in the conversation.
Do not confuse with
| Skill / process | Difference |
|---|
Bundled /review | Orchestrator + scratch review files / PR pending comments; still apply this bar if used for correctness, and land follow-ups in docs |
code-review | Maintainability / code judo / file size |
| Pre-ship gate | Review → fix → re-review before Done (Agents.md, pr1-uncommitted-review-gate.md) |
One-liner (if protocol path is forgotten)
Read docs/agent/phenomenal-review.md and execute it against local changes.
Assume the code is wrong until evidence says otherwise.
Write findings and follow-up tasks into the docs as the protocol requires.