review-code-you-did-not-write

A review procedure for checking code written by someone else, including code made by an AI agent, before it is merged or released.

In plain words
What is it for?
Use it to review pull requests, dependency updates, templates, snippets, or agent-written changes that you do not fully understand before approving them.
Why use it?
It helps you find mistakes that can look correct when you only read a summary or see passing tests. It gives you a way to judge code without relying on the original author's intent.

Skill for Claude CodeCodex

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 skills/everywan-dev/claude-code-engineering/review-code-you-did-not-write
Any agent
npx skills add everywan-dev/claude-code-engineering --skill review-code-you-did-not-write
Clone the repo
git clone --depth 1 https://github.com/everywan-dev/claude-code-engineering

Made for: Claude Code, Codex.

Per session 52 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,255 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 $0.00052 $0.01255
Opus 5 $0.00026 $0.00628
Sonnet 5 $0.00010 $0.00251
Haiku 4.5 $0.00005 $0.00126

Measured 2d ago against content hash 781948d42b5a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

review-code-you-did-not-write 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.

skills/review-code-you-did-not-write/SKILL.md · 105 lines

How it starts

The opening of the file, as written. The whole thing — 105 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Review code you did not write

Reviewing your own code is hard because you know what you meant. Reviewing code you did not write is hard for the opposite reason: you have no intent to compare against, so the only thing left to judge is whether it looks right.

Looking right is what generated code is best at. It has the shape of correct code, the naming of correct code and the confidence of correct code. That is not a criticism of the generator; it is the job description.

🔴 You are not reviewing whether the code is plausible. You are reviewing whether you could have caught it being wrong. If nothing in your review could have produced a rejection, you did not review it — you accompanied it.

When to use this

  • An agent produced a diff and you are about to merge it
  • A dependency, snippet or template arrived from outside and is now yours
  • You are approving a pull request in an area you do not own
  • Something works and you cannot explain why it works
  • You are about to write "LGTM" on more lines than you have read

The procedure

1. Read the diff, not the summary. The summary is written by the same party that wrote the code and shares its blind spots. If the summary and the diff disagree, the diff is the truth — and the disagreement is itself the finding.

2. For each change, ask what it would look like if it were wrong. Not "is this right?" — that question has an easy yes. Ask what a broken version of this would look like, then check whether you would be able to tell the difference. Where you cannot tell, you have found the part that needs a real check.

3. Run the thing the change claims to fix, in the state a user would. Not the test suite: the actual claim. A change that "fixes the install" is verified by installing, from a clean state, using the documented command, and nothing else counts.

4. Hunt for the confident no-op. The most dangerous generated change is the one that runs cleanly and does nothing:

Shape How it reads What it does
A guard that never triggers Defensive Dead code that hides the real path
A check comparing a value to itself Thorough Passes by construction
An exception handler that swallows Robust Turns a failure into silence
A config written where nothing reads it Configurable No effect at all
A test asserting on data it just built Well tested Cannot disagree with the code

Read the full file on GitHub · 105 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 · 105 lines · 52 tokens per session scan A 781948d42b5a

Subscribe to this mod's changes

review-code-you-did-not-write is a skill published in the GitHub repository everywan-dev/claude-code-engineering (2 stars, last pushed 13d ago), licensed Apache-2.0. It adds 52 tokens to every session and 1,255 once invoked, about $0.0003 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.