/skill-code-review

name: skill-code-review
description: "Expert multi-AI code review with inline PR comments — use for thorough quality and security analysis"
disable-model-invocation: true

> **Host: Codex CLI** — This skill was designed for Claude Code and adapted for Codex. > Cross-reference commands use installed skill names in Codex rather than `/octo:*` slash commands. > Use the active Codex shell and subagent tools. Do not claim a provider, model, or host subagent is available until the current session exposes it. > For host tool equivalents, see `skills/blocks/codex-host-adapter.md`.

Code Review Skill

MANDATORY COMPLIANCE — DO NOT SKIP

**When this skill is invoked, you MUST execute the multi-LLM review pipeline. You are PROHIBITED from:**

• Doing a direct single-model code review without multi-provider synthesis
• Deciding the scope is "too broad" and narrowing it without asking the user
• Skipping the provider check or structured review phases
• Substituting two background Sonnet agents for the full multi-provider pipeline
• Rationalizing "a focused audit would be more effective" — the user wants multi-LLM perspectives

**Your first output line MUST be:** `🐙 **CLAUDE OCTOPUS ACTIVATED** - Multi-LLM Code Review`

Invokes the code-reviewer persona for thorough code analysis during the `ink` (deliver) phase.

Quick Mode

For fast sanity checks (staged changes, small PRs), skip the full review pipeline and run just two phases:

# Quick: grasp (consensus on scope) → tangle (parallel review)
"${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-${HOME}/.claude-octopus/plugin}}/scripts/orchestrate.sh" grasp "[review request]"
"${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-${HOME}/.claude-octopus/plugin}}/scripts/orchestrate.sh" tangle "[synthesized scope]"

Use quick mode when user says "check this PR", "quick review", "sanity check my changes", or for pre-commit checks. Use the full review for PRs with security/architecture impact.

Usage

# Via orchestrate.sh
"${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-${HOME}/.claude-octopus/plugin}}/scripts/orchestrate.sh" spawn code-reviewer "Review this pull request for security issues"

# Via auto-routing (detects review intent)
"${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-${HOME}/.claude-octopus/plugin}}/scripts/orchestrate.sh" auto "review the authentication implementation"

Capabilities

• AI-powered code quality analysis
• Security vulnerability detection
• Performance optimization suggestions
• Architecture and design pattern review
• TDD compliance and test-first evidence review
• Autonomous code generation risk detection
• Best practices enforcement

Persona Reference

This skill wraps the `code-reviewer` persona defined in:

• `agents/personas/code-reviewer.md`
• CLI: `codex-review`
• Model: `gpt-5.2-codex`
• Phases: `ink`

Example Prompts

"Review this PR for OWASP Top 10 vulnerabilities"
"Analyze the error handling in src/api/"
"Check for memory leaks in the connection pool"
"Review the test coverage for the auth module"

Autonomous Implementation Review

When the review context indicates `AI-assisted`, `Autonomous / Dark Factory`, or unclear provenance, raise the rigor bar. Do not treat generated code as trustworthy just because it is polished.

TDD Evidence

Check for concrete signs that the change followed red-green-refactor rather than test-after implementation:

• Compare the diff and recent history when available to see whether tests were added before or alongside production changes.
• Prefer behavior-defining tests over snapshot-only or mock-heavy tests that merely restate the implementation.
• Verify the production code looks like the minimum needed to satisfy the tests, rather than a speculative abstraction with unused options.
• If evidence is missing, mark TDD compliance as unknown and do not assume TDD happened.

Autonomous Codegen Risk Patterns

Elevate or add findings when you see patterns common in high-autonomy output:

• Option-heavy APIs or abstractions not justified by tests or current requirements
• Placeholder logic, TODO/FIXME-driven control flow, or dead branches that appear "future ready"
• Mock, fake, or dummy behavior leaking into production paths
• Unwired components, unused helpers, or code that exists without an execution path
• Silent failure handling, broad catch blocks, missing logs, or weak operational visibility
• Missing rollback notes, migration guards, or release-safety checks for risky changes

Review Output Addendum

Add a short section to the review synthesis when autonomy or TDD is in scope:

## TDD / Autonomy Assessment

- Provenance: Human-authored | AI-assisted | Autonomous / Dark Factory | Unknown
- TDD evidence: Confirmed | Partial | Unknown
- Autonomous risk signals: None | Minor | Significant
- Recommendation: Ship | Fix before merge | Re-run with /octo:tdd or tighter supervision

Implementation Completeness Verification

After the code-reviewer persona completes, run stub detection to verify implementation completeness.

Stub Detection Process

**Step 1: Get changed files**

# Get files changed in the commit/PR
if [ -n "$COMMIT_RANGE" ]; then
    changed_files=$(git diff --name-only "$COMMIT_RANGE")
else
    changed_files=$(git diff --name-only HEAD~1..HEAD)
fi

# Filter for source code files
source_files=$(echo "$changed_files" | grep -E "\.(ts|tsx|js|jsx|py|go)$")

**Step 2: Check for stub patterns**

For each changed file, check for common stub indicators:

for file in $source_files; do
    echo "Checking $file for stubs..."

    # Check 1: Comment-based stubs
    stub_count=$(grep -E "(TODO|FIXME|PLACEHOLDER|XXX)" "$file" 2>/dev/null | wc -l | tr -d ' ')

    if [ "$stub_count" -gt 0 ]; then
        echo "⚠️  WARNING: Found $stub_count stub indicators in $file"
        grep -n -E "(TODO|FIXME|PLACEHOLDER)" "$file" | head -3
    fi

    # Check 2: Empty function bodies
    empty_functions=$(gre