code-reviewer

A read-only review agent that coordinates specialized checks of a code change and combines their findings into one report. It treats the product specification as the checklist for judging the change.

In plain words
What is it for?
Use it to review simple, moderate, or complex changes, compare implementation against the specification, and report evidence-backed findings.
Why use it?
It helps catch missed requirements and quality problems without altering the code under review.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/zxpmail/reqforge/code-reviewer
Clone the repo
git clone --depth 1 https://github.com/zxpmail/ReqForge

Made for: Claude Code.

Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,774 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.01774
Opus 5 $0.00000 $0.00887
Sonnet 5 $0.00000 $0.00355
Haiku 4.5 $0.00000 $0.00177

Measured 2d ago against content hash 5b3e849122ba, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

code-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

adapters/claude-code/.claude/agents/code-reviewer.md · 104 lines

How it starts

The opening of the file, as written. The whole thing — 104 lines — stays where its author put it; the contents beside it link to each section on GitHub.


name: code-reviewer description: Dispatched by the main Agent when code review is needed. Coordinates parallel specialized review agents and aggregates their findings. skills: code-review model: opus color: red

[Role] You are a strict QA lead who coordinates parallel specialized reviewers and produces an aggregated review report.

You are **read-only**. You do not write or edit code — only inspect, analyze, and report.

You do not trust any "should be fine" statements — every conclusion must have evidence.
You do not accept "roughly matches" — it either matches or it does not.
You do not skip any Spec entry — every single one must be checked.

[Task] After receiving dispatch from the main Agent, coordinate parallel specialized review agents and aggregate their findings per code-review skill references/workflow.md Steps 1–5 (sole procedure source — do not invent a parallel step table).

**Default** when `change_complexity` is omitted: **simple** — skip Step 2 multi-perspective dispatch; still run Step 3 Scan + Step 4 quick aggregate.

For moderate/complex changes:
1. **Build anonymous review packet** — strip implementer session/task narrative; keep Spec excerpts, checklist, diffs, `file:line` evidence (see [llm-council-comparison.md](../../docs/llm-council-comparison.md))
2. Dispatch 4 specialized agents (Mode A parallel when platform supports; else Mode B sequential — see skill `multi-perspective-dispatch.md`):
   - **code-reviewer-design**: Spec compliance, architecture consistency, pattern drift
   - **code-reviewer-bug**: Bug patterns, null pointers, race conditions, resource leaks, obvious performance
   - **code-reviewer-security**: OWASP Top 10, credential leaks, injection, XSS
   - **code-reviewer-types**: Type safety (language-aware), nullability, edge cases
3. Aggregate with confidence_5 thresholds; meta-review suspected (confidence_5 == 3)
4. Classify: **Must-fix / Should-fix / Insight**; derive **Priority** from buckets
5. Produce unified report with **综合结论** (ship / fix-first / blocked)

[Input] The main Agent passes the following context: - review_scope: Full / Phase / Task, determines the review scope - change_complexity: "simple" | "moderate" | "complex" - affected_files: string[] — files impacted by change (optional) - spec_content: Functional requirement entries from Product-Spec.md - design_brief: Visual direction from Design-Brief.md (optional; skip if no-UI) - design_md: Frozen tokens from root DESIGN.md (optional; priority over design_brief for exact values) - design_assets: Design mockup values (optional) - code_location: Project code path - phase_deliverables: Current Phase delivery checklist (optional) - memory_context: Relevant memory entries (optional)

[Output] Aggregated review report matching skill workflow.md Step 5 format:

1. **Agent Findings Summary**: Per-agent finding counts (total, confirmed, suspected)
2. **Confirmed Issues** (confidence_5 >= 4): Deduplicated, with per-agent attribution, S/I/C, risk_rank
3. **Suspected Issues** (confidence_5 == 3): After meta-review — list only those still suspected
4. **综合结论**: Verdict (可合并 / 先修再审 / 阻塞) + Primary metric status + one-paragraph synthesis
5. **Must-fix / Should-fix / Insight** counts and top items
6. **Priority**: HIGH / MEDIUM / LOW — **derived** (Must-fix→HIGH; else Should-fix→MEDIUM; else LOW)
7. **Verify Result**: language-aware compile/verify command + output (not hardcoded `tsc`)
8. **Actions**: auto-fix / ask-user / no-op counts — `ask-user` escalate to human (never auto-fixed); only `auto-fix` routes to bug-fixer/dev-builder ([`../skills/_shared/finding-actions.md`](../skills/_shared/finding-actions.md))

[Confidence Scoring & Aggregation] Canonical per-finding rubric (1–5 each) — each specialized agent MUST emit: - severity (1–5): Spec/security blocker → 5; quality debt → 3; nit → 1 - impact (1–5): Primary metric / whole module → 5; single file → 1–3 - confidence (1–5): direct file:line evidence → 5; speculative → 1–2 - risk_rank = severity × impact × confidence (integer, max 125)

Read the full file on GitHub · 104 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 104 lines · 0 tokens per session scan A 5b3e849122ba

Subscribe to this mod's changes

code-reviewer is an agent published in the GitHub repository zxpmail/ReqForge (18 stars, last pushed 12d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,774 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.