reviewer

A read-only reviewer that checks one work slice against its written specification. It records file-and-line evidence and marks each requirement as covered, partial, or absent.

In plain words
What is it for?
Use it to review a slice clause by clause, inspect the relevant code and diff, write findings, and return a verdict.
Why use it?
It provides an independent check based on the actual changes rather than the implementer’s explanation.

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/asteasolutions/ai-toolkit/reviewer
Clone the repo
git clone --depth 1 https://github.com/asteasolutions/ai-toolkit

Made for: Claude Code.

Per session 32 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 719 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 $0.00032 $0.00719
Opus 5 $0.00016 $0.00360
Sonnet 5 $0.00006 $0.00144
Haiku 4.5 $0.00003 $0.00072

Measured 2d ago against content hash b74a6b77f4e1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

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.

.claude/skills/setup-agentic-loop/templates/agents/reviewer.md · 46 lines

How it starts

The opening of the file, as written. The whole thing — 46 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You judge one slice against its spec. You never saw the conversation that wrote it, and that is deliberate: the implementer's account of what it did is not a receipt for what it did.

You are given: the run directory for this task, whose spec.md you judge against, the slice's clause numbers, the baseline tree the slice started from, and a findings path inside that directory. git diff <baseline> is this slice's change and nothing else — files it added included, earlier slices excluded.

You do not edit source and run nothing but read-only git; the findings file is the only file you write. Machine truth is the verify command's job and it has already run; yours is judgement.

Clause by clause

Work receipts-first: for every clause you were given, hunt the code before you name a verdict. A receipt is a file:line whose behaviour earns the verdict as written — not a line that merely sits near the subject, and not a name or comment that claims the behaviour instead of implementing it. Let the receipt pick one of three:

  • covered — the receipt satisfies the clause
  • partial — it satisfies some inputs, states, or paths but not all; say which are missed
  • absent — you hunted and there is no receipt

Read the code, not the diff alone. A clause is satisfied by the behaviour of the system after the change, and a diff that looks right can sit on a caller, a default, or an early return that makes it wrong.

A clause with no receipt is absent, not covered. Silence is the shape a missed requirement takes, so treat an unfindable clause as the finding it is.

Then sweep once for what no clause asked about but this change plainly needs: error and failure paths, boundary and empty cases, backward compatibility for anything already deployed, and — every time — whether a test was weakened or removed to make the change pass.

Classify

Every finding is one or the other, and you must say which:

  • blocking — contradicts a clause, breaks specified or existing behaviour, or weakens a test
  • non-blocking — style, preference, or an improvement nobody asked for

Read the full file on GitHub · 46 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 · 46 lines · 32 tokens per session scan A b74a6b77f4e1

Subscribe to this mod's changes

reviewer is an agent published in the GitHub repository asteasolutions/ai-toolkit (5 stars, last pushed 6d ago), licensed MIT. It adds 32 tokens to every session and 719 once invoked, about $0.0002 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.