phenomenal-review

Adversarial correctness and contract review of a git diff via docs/agent/phenomenal-review.md (split-context reviewer stance; optional multi-pass). Produces structured findings and checkable follow-up tasks under docs/ (not chat-only). Use when the user asks for a phenomenal review, deep review, adversarial review, contract review, correctness review of local/branch/PR changes, or wants residual work filed into plan docs. Do not use for pre-ship fix loops, maintainability/code-judo audits, or implementing fixes unless the user explicitly asks to harden after the review.

Jump to install

Repository: scoizzle/Poly
Last source activity: July 31, 2026 at 02:44
Detected SKILL.md language: English
Stars: 5
Forks: 1

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/scoizzle/Poly --skill phenomenal-review

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

SKILL.md

Source instructions · Read-only preview

Back to search Back to installation

Phenomenal review (Copilot / Agent Skills wrapper)

This skill is a thin adapter. The full protocol is tool-agnostic and lives in the repository:

docs/agent/phenomenal-review.md

Also see: docs/agent/README.md.

Instructions

Open and follow docs/agent/phenomenal-review.md completely. Do not substitute a lighter checklist.

Determine the review target (default: uncommitted local changes; or branch / PR / paths if specified) and mode (standard or multi).

Adversarial stance: assume the code is wrong; reviewer is not implementer; prefer diff-first. If this session wrote the code, use multi-pass with a fresh context and the Pass B template (docs/agent/phenomenal-review.md §3.7.1).

Collect the diff, trace producer/consumer call graphs, hunt fail-closed and contract bugs (sibling-path drift, reachability before severity, primary-evidence counts, invariant comments as checklists, same-shape-different-meaning, oracle weakening), audit tests and plan honesty — as the protocol specifies.

Do not modify production or test code unless the user explicitly requests a harden/pre-ship pass after the review.

Required deliverables:

A structured review markdown file under docs/ (suite path or docs/agent/reviews/YYYY-MM-DD-<slug>.md).
Checkable follow-up tasks under docs/ (including process fixes when bug classes recur). Chat alone is not enough.

End with a short user report: verdict, bug/suggestion/nit counts, paths to the two docs, top issues.

Related (do not merge)

Always-on repo rules: AGENTS.md / Agents.md, .github/copilot-instructions.md

Pre-ship fix gate: docs/plans/v2-to-v3/simple-agent-tasks/pr1-uncommitted-review-gate.md

Maintainability audits: separate skill if present; this skill is correctness/contracts first