Nc9 Skills Review

Review

Local, high-signal code review. The skill builds a bundle of context on disk, hands it to Codex via MCP, and renders the structured JSON verdict back to the user.

When to Use

• User says "review", "/review", "review my changes".
• Before committing non-trivial work.
• Before opening a PR.
• After addressing review feedback — to confirm resolution.

Workflow

1. Assemble the bundle

Run prepare to gather disk-side context (git diff, issues, plan, conventions). It prints the bundle directory path as the last line of stdout.

./scripts/review prepare \
  --issues "#123,PROJ-456" \
  --plan ./plan.md \
  --title "Refactor auth flow"

prepare writes:

• CHANGES.diff — the unified diff
• CHANGED_FILES.txt — touched paths
• ISSUES.md — fetched GitHub / Linear / Sentry context
• PLAN.md — --plan argument or most recent ~/.claude/plans/*.md
• CONVENTIONS.md — repo REVIEW.md + CLAUDE.md + user ~/.claude/CLAUDE.md
• REFERENCED_FILES.md — contents of --files paths
• AGENT_CONTEXT.md — placeholder; you fill this in next
• REVIEW_PROMPT.md — the full reviewer prompt (copied from this skill)
• MANIFEST.json — file list + metadata

Default source: uncommitted changes if any exist, otherwise prompts for a base branch or commit.

2. Write AGENT_CONTEXT.md

Only you (the agent) have the conversation context. Overwrite the placeholder <bundle>/AGENT_CONTEXT.md with what's not on disk:

• User request — quote the original ask and any clarifying turns.
• User feedback — corrections, preferences, course-corrections during the work.
• Plan iteration — key decisions and tradeoffs that shaped the change, especially things not captured in the plan file.
• Task summary — one-paragraph description of what was built and why.

Be concrete. This is the canonical statement of intent the reviewer uses for spec-conformance judgments.

3. Invoke Codex via MCP

Read <bundle>/REVIEW_PROMPT.md and pass its contents to mcp__codex__codex along with the bundle path. Recommended config:

{
  "approval-policy": "never",
  "sandbox": "workspace-write"
}

Prompt shape:

{REVIEW_PROMPT.md contents}

Review bundle directory: <absolute bundle path>

Read every file in that directory (check MANIFEST.json for the list) before producing your JSON response.

Capture the returned threadId from structuredContent.threadId.

4. Verification pass (required)

Call mcp__codex__codex-reply with the same threadId:

For each issue you returned, re-read the file and lines cited in evidence. Drop any finding where the citation does not substantiate the claim (wrong file, wrong line, code doesn't match the description, or you inferred from naming rather than reading). Return the updated JSON object — same schema, only verified findings. Recompute severity_counts and score accordingly.

This single step is the biggest quality lever — it suppresses hallucinated findings that name functions or lines that don't exist.

5. Parse, persist, render

• Parse the final JSON. If Codex wrapped it in prose or code fences, strip them and retry parsing; if it still doesn't parse, send one more codex-reply asking for valid JSON only.
• Write the verified JSON to <bundle>/review.json.
• Render a human summary to the user:
  • Score + verdict headline (e.g. 4/5 — ready-with-nits).
  • One-line summary.
  • spec_alignment paragraph.
  • Issues grouped by severity (critical → high → medium → nit → pre_existing), each with title, files[].path:lines, and one-line description. Deep-dive details available in review.json.
  • strengths and improvements bullets if present.
  • Path to <bundle> so the user can inspect raw artifacts.

Keep the rendered output terse — the user can read review.json for detail.

Fallback: Codex MCP unavailable

If mcp__codex__codex isn't loaded in the session, fall back to the Codex CLI with the same bundle:

codex exec \
  --cd <bundle path> \
  --sandbox workspace-write \
  --ask-for-approval never \
  "$(cat <bundle>/REVIEW_PROMPT.md)

Review bundle directory: <bundle path>
Read every file in that directory before producing your JSON response."

Then run a second codex exec against the same working directory for the verification pass, referencing the JSON written in step 1. Parse and render as in step 5.

Output schema

See REVIEW_PROMPT.md for the full schema. Summary:

{
  "score": 0,
  "verdict": "ship | ready-with-nits | needs-changes | significant-changes | rethink",
  "summary": "...",
  "severity_counts": { "critical": 0, "high": 0, "medium": 0, "nit": 0, "pre_existing": 0 },
  "spec_alignment": "...",
  "issues": [
    {
      "id": "R-001",
      "severity": "critical | high | medium | nit | pre_existing",
      "category": "bug | security | spec | performance | maintainability | tests",
      "title": "...",
      "description": "...",
      "files": [{ "path": "...", "lines": "..." }],
      "suggested_fix": "...",
      "confidence": "high | medium | low",
      "evidence": "file:line — justification"
    }
  ],
  "strengths": [],
  "improvements": []
}

Verdict ladder

Command

./scripts/review prepare [OPTIONS]

Options

Requirements

• mcp__codex__codex + mcp__codex__codex-reply tools loaded (preferred), or codex CLI installed and authenticated (fallback).
• Codex MCP server: set client_session_timeout_seconds high (e.g. 360000) in your MCP client config — reviews routinely run many minutes.
• gh CLI for GitHub issues.
• LINEAR_API_KEY (optional) for Linear issues.
• SENTRY_AUTH_TOKEN, SENTRY_ORG (optional) for Sentry issues.

Per-repo customization — REVIEW.md

Drop a REVIEW.md at the repo root to inject house rules (skip paths, nit caps, mandatory checks). It concatenates into CONVENTIONS.md in every bundle and overrides the generic prompt where they conflict. See REVIEW.md.example in this skill for a template.

Examples

# Most common: review uncommitted changes, auto-pick latest plan
./scripts/review prepare

# Explicitly review staged/unstaged, include linked issue
./scripts/review prepare --uncommitted --issues "#142"

# Review the last commit
./scripts/review prepare --commit HEAD

# Branch-diff review against main
./scripts/review prepare --base main

Source: nc9/skills — distributed by TomeVault.