Codex Code Review

Trigger

• Keywords: review, PR, code review, second opinion, audit, check

When NOT to Use

• Document review (use doc-review)
• Security-specific review (use security-review)
• Test coverage review (use test-review)
• Just want to understand code (use code-explore)

Variants

Shared Workflow

Step 0 (PENDING) → Collect changes → [Pre-checks if Full] → Dual Review (Codex + Task) → Await Results → Aggregate → Emit Gate → Loop if Blocked

Step 0: Dual Review Init (Fail-closed)

Execute: bash scripts/emit-review-gate.sh PENDING

This sets review_mode=dual and aggregate_gate.executed=false in state file, ensuring fail-closed semantics — if the process crashes before Step 4.5, stop-guard blocks.

Step 1: Collect Change Metadata

Collect metadata only — Codex reads the actual diffs and file contents itself via sandbox access.

Codex independently reads full diffs and file contents via git diff HEAD -- <file> + cat (per research instructions).

Step 1.5: Feature Context & AC Detection (Spec-Driven Review)

Execute: bash scripts/resolve-feature.sh → parse JSON output.

If has_requests=true AND confidence in (high, medium):

1. Glob ${docs_path}/requests/*.md, sort descending, take latest
2. Read latest request doc
3. Extract ## Acceptance Criteria section (parse - [ ] / - [x] items)
4. Filter out quality-gate ACs matching: /codex-review-fast, /codex-review-doc, /codex-review, /precommit, /precommit-fast, /pr-review
5. Cap: max 20 ACs (truncate with "... and N more" note)
6. Build SPEC_CHECKLIST variable, set REQUEST_DOC_PATH

Graceful degradation: resolve-feature fails / no requests / no AC section / parse error → SPEC_CHECKLIST = null (skip silently).

Step 1.6: Deferred Finding Context

Prerequisite: Requires R4 (Nit History Persistence) hook-side write path. Until R4 is deployed, .claude_nit_history.json will not exist and this step is a no-op (graceful degradation).

Read .claude_nit_history.json (if exists):

1. Filter .deferred[] entries where last_seen + ttl_days > now
2. Sanitize each entry before injection (mandatory):
   • canonical_issue <= 120 chars (truncate)
   • Strip markdown control chars (**, backticks, #, >, |)
   • No raw code snippets; only file:line references
   • No secrets/tokens/passwords/API keys (per rules/security.md)
   • Reject entries with shell metacharacters (;, &, |, backtick, $()
3. Format as <deferred_context> XML block (max 10 entries)
4. Store in DEFERRED_CONTEXT variable

Inject into all prompt variants after research instructions, before Review Dimensions:

${DEFERRED_CONTEXT ? DEFERRED_CONTEXT : ''}

Format:

<deferred_context>
Previously deferred (do not re-report without new evidence):
- [Nit] src/service.ts | naming convention (deferred 2x)
- [P2] src/utils.ts | error handling pattern (deferred 1x)
</deferred_context>

Graceful degradation: file missing / invalid JSON / no entries / sanitization failure → DEFERRED_CONTEXT = null (skip silently).

Step 2: Pre-checks (Full variant only)

{LINT_FIX_COMMAND}
{BUILD_COMMAND}

These placeholders are resolved from the host project's CLAUDE.md or package.json scripts. Record results as LOCAL_CHECKS.

Step 3: Dual Review (Parallel Dispatch)

Case A: First review (no --continue)

Launch two reviewers in parallel (single message, multiple tool calls):

1. Codex MCP (primary): Use mcp__codex__codex with variant-specific prompt:

   Variant	Prompt Template
   Fast	references/codex-prompt-fast.md
   Full	references/codex-prompt-full.md
   Branch	references/codex-prompt-branch.md

   Config: sandbox: 'read-only', approval-policy: 'never'

   Save the returned threadId.

2. Secondary reviewer: Use Task tool with reviewer selection cascade:

   Priority	Reviewer	subagent_type	Condition
   1	pr-review-toolkit:code-reviewer	pr-review-toolkit:code-reviewer	Default choice
   2	strict-reviewer	strict-reviewer	Priority 1 fails/times out
   3	Codex-only (degraded)	—	Both unavailable

   Selection: Try priority 1 first. If Task fails or times out (30s), try priority 2. If both unavailable, fall back to Codex-only (degraded mode — proceed with Codex results only, apply degradation matrix from references/review-common.md).

   Task prompt (provide changed file list + diff stats, request P0/P1/P2/Nit findings in standard output format):

   Review the code changes for correctness, security, performance, and maintainability issues.
   
   ## Changed Files
   <git diff --name-only output>
   
   ## Diff Stats
   <git diff --stat output>
   
   Read the actual diffs and file contents yourself to perform the review.
   
   Before reporting findings, independently verify each one:
   1. Evidence check: what specific code proves it's real? (file:line)
   2. Context check: did you read enough surrounding code?
   3. False positive check: could it be intentional design?
   4. Severity check: could it be more severe than initially assessed?
   5. Gap check: what related issues might you have overlooked?
   Only report findings that survive all 5 checks.
   
   Output findings in this format:
   - [P0/P1/P2/Nit] file:line issue description → fix recommendation
   
   Group by severity. Include a final gate: ✅ Ready (no P0/P1) or ⛔ Blocked (has P0/P1).
   

Case B: Loop review (has --continue)

• Codex: Use mcp__codex__codex-reply with re-review template from references/review-common.md
• Secondary: Re-dispatch in parallel (same mechanism as first pass, fresh context). Always dispatched in v1 — no skip exception. Cycle resets on any code edit.

Step 3.5: Await Codex + Reconcile Secondary

Codex is the blocking reviewer — await its result for the initial gate. Secondary runs in background (run_in_background: true) and is non-blocking:

Step 4: Consolidate Output (Dual Mode)

1. Normalize both sets of findings to unified format: [severity] file:line description → fix

   • Codex findings: already in standard format
   • toolkit findings: apply Severity Mapping (see references/review-common.md § Severity Mapping)
   • strict-reviewer findings: already use P0/P1/P2/Nit

2. Deduplicate using key = file + canonical_issue_text (ignore line ±5 difference)

   • Same key → keep highest severity (P0 > P1 > P2 > Nit)

3. **T