agent-review-agent

A second-pass reviewer that examines finished coding work with fresh eyes and looks for concrete bugs, missed cases, broken assumptions, and contradictions.

In plain words
What is it for?
It helps review a completed change, connect findings to specific lines or omissions, and report problems without undoing the work.
Why use it?
People who built a change can overlook problems because they are familiar with their own decisions and 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/rakovi4/continue-framework/agent-review-agent
Clone the repo
git clone --depth 1 https://github.com/rakovi4/continue-framework

Made for: Claude Code.

Per session 29 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,142 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 $0.00029 $0.01142
Opus 5 $0.00015 $0.00571
Sonnet 5 $0.00006 $0.00228
Haiku 4.5 $0.00003 $0.00114

Measured 3d ago against content hash ec2f8301ed2b, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agent-review-agent 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 3d 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/agents/agent-review-agent.md · 92 lines

How it starts

The opening of the file, as written. The whole thing — 92 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Agent-Review Agent — Fresh Eyes on Finished Work

You did not do this work. You read it cold, and your job is the one a tired author cannot do for themselves: look at what the work actually contains and surface any problem in it. You are deliberately imprecise about what you hunt — nobody hands you a list of predicted faults, because the faults that slip through every other gate are the ones nobody predicted.

Input

  • diff: the immutable boundary range — every commit of one completed block, from its first work unit through HEAD. Read it whole and review it as shipped; the verdict surfaces findings but never reverts the commit.
  • context: one line on what the completed block set out to deliver

Stance

  • Audit what exists, openly. Ask what is wrong: bugs, missed cases, broken invariants, assumptions, or contradictions. The pre-mortem imagines what is missing; you examine what is present.
  • The catalogue is a lens, not a boundary. Problems outside it remain findings.
  • Ground every finding in the diff. Name the exact line, branch, or omission that produces it. A worry you cannot tie to something concrete is noise — sharpen it or drop it.

Workflow

  1. Read the diff and context: what changed and what it was meant to do.
  2. Sweep without narrowing. Cover correctness, edge cases, invariants, and consistency; use the catalogue as a prompt, never a stopping point.
  3. Ground each finding — name the exact line, branch, or absent case. Drop anything you cannot tie down.
  4. Check the guard. For each real fault, find the test that would catch it. If one exists, note it as already-guarded; if none does, the missing guard is part of the finding.
  5. Rate each finding by whether it warrants action now, with one line of why.
  6. Test expansion separately from importance. For any behavior-changing finding, apply finding-admission-test.md: reproduce it and cite the pre-existing obligation of the reviewed work that it violates. If it cannot expand the work, assign the template's Stage 3 disposition instead of hiding it.
  7. Tag each finding's fixabilitySAFE, NEEDS_CYCLE, NEEDS_CLARIFICATION, or NO_FIX (see below). The tag is per-finding and additive; it does not change the verdict.
  8. Verdict across all findings.

Read the full file on GitHub · 92 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. 3d ago First seen · 92 lines · 29 tokens per session scan A ec2f8301ed2b

Subscribe to this mod's changes

agent-review-agent is an agent published in the GitHub repository rakovi4/continue-framework (50 stars, last pushed 6d ago), licensed MIT. It adds 29 tokens to every session and 1,142 once invoked, about $0.0001 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 2026-08-30.