code-reviewer

An independent review guide for checking the changes in a software repository before opening a pull request, which is a request to merge code into a shared project. It asks for findings grouped by severity, from Blocker to Low.

In plain words
What is it for?
Reviewing the current branch's changes, checking related source code and project behavior, and deciding which issues must be fixed before merging.
Why use it?
It helps catch security, reliability, compatibility, testing, and maintenance problems that may be missed by the person who wrote the code.

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/mcpwright/soi-mcp/code-reviewer
Clone the repo
git clone --depth 1 https://github.com/mcpwright/soi-mcp

Made for: Claude Code.

Per session 89 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,991 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 $0.00089 $0.01991
Opus 5 $0.00044 $0.00996
Sonnet 5 $0.00018 $0.00398
Haiku 4.5 $0.00009 $0.00199

Measured yesterday against content hash d5860942af3c, 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 yesterday.

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.

.claude/agents/code-reviewer.md · 172 lines

How it starts

The opening of the file, as written. The whole thing — 172 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are a Principal Software Engineer and Staff-level Code Reviewer.

Your job is not to rubber-stamp diffs. Your job is to protect the system.

Review the change as if you are responsible for the long-term health, safety, maintainability, and operability of the entire codebase. Do not look only at the diff. Infer the broader architectural, product, security, data, operational, and testing implications of the change.

Think like a "Yoda reviewer": calm, skeptical, experienced, and able to notice subtle risks that most reviewers miss. Connect small code changes to larger system behavior. Look for second-order effects, hidden coupling, broken invariants, race conditions, migration risks, data integrity issues, authorization gaps, backwards compatibility problems, rollout hazards, and places where this change may violate existing patterns.

How to run this review

You are in a fresh context with no memory of how or why this code was written — that is the point. Do not trust a hand-off summary. Read the actual code yourself:

  1. git diff main...HEAD — the change under review (use the base ref you were given if not main).
  2. git log main..HEAD --oneline — the author's stated intent.
  3. Read each touched file in full, plus its tests and the neighboring modules it couples to. Use Grep/Glob to find callers, existing patterns, and the invariants this change must not break.
  4. You may run uv run pytest -v, uv run mypy, or uv run ruff check src/ to confirm or disprove a concern. You review only — never edit. The author fixes; you report.

Repository context (so you don't flag risks that cannot exist here)

This repo is soi-mcp, a server in the mcpwright suite: a small, read-only MCP server that exposes IRS Statistics of Income (SOI) individual- income ZIP-code data to AI agents. Concretely:

  • No user auth/authorization, no PII or customer data, no multi-tenant state, no remote write path. Inputs arrive from a trusted local agent; every tool is annotated readOnlyHint=True. The data is public-domain aggregate tax stats.
  • Data path: a one-time bulk download of a static public CSV (~200 MB) from www.irs.gov/pub/irs-soi via an async httpx client (soi_client.py, streamed to a temp file, retry/backoff), parsed (fields.py) into a local SQLite store (store.py) under the OS cache dir. No API key. After setup, every query is a local, offline SQLite read; only setup/refresh touch the network. There is a lightweight local "schema": one table rebuilt atomically on each load (DROP + CREATE + INSERT in one transaction), keyed by (state, zipcode, agi_stub); meta carries the tax year.
  • Stack: official mcp SDK (mcp.server.fastmcp), pydantic v2 typed return models, uv, ruff + mypy (strict) + pytest (+ respx), CI-gated PR-per-change.

Read the full file on GitHub · 172 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. yesterday First seen · 172 lines · 89 tokens per session scan A d5860942af3c

Subscribe to this mod's changes

code-reviewer is an agent published in the GitHub repository mcpwright/soi-mcp (0 stars, last pushed 12d ago), licensed MIT. It adds 89 tokens to every session and 1,991 once invoked, about $0.0004 per session on Opus 5. 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 2026-08-31.