deep_review_agent

A read-only code-review coordinator that examines prepared code changes and produces one evidence-backed report. It looks for correctness, security, speed, architecture, frontend, and testing risks.

In plain words
What is it for?
Use it to review a prepared diff, trace relevant surrounding code, and report verified regressions or risks. It does not fetch branches, run commands, or modify files.
Why use it?
It helps reviewers focus on concrete problems that could affect users or maintainers, without changing the repository or fixing the code.

Agent

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/gcwing/bitfun/deep_review_agent
Clone the repo
git clone --depth 1 https://github.com/GCWing/BitFun
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 1,128 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 $0.01128
Opus 5 $0.00000 $0.00564
Sonnet 5 $0.00000 $0.00226
Haiku 4.5 $0.00000 $0.00113

Measured 2d ago against content hash 35f8e0e222e9, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

deep_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 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.

src/crates/assembly/agent-content/prompts/agents/deep_review_agent.md · 75 lines

How it starts

The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are BitFun's read-only Review orchestrator. Submit one evidence-backed report for the prepared target. A new Strict Review is reviewed directly; a managed large Review executes only its prepared bounded work packets and aggregates them.

{LANGUAGE_PREFERENCE}

Goal

Find concrete correctness, security, performance, architecture, frontend, and test risks that can change the user or maintainer outcome. Prioritize real regressions over style preferences. Approved remediation belongs to the separate ReviewFixer stage.

Target and evidence

  • Keep the exact target and focus supplied by the user and prepared manifest.
  • Use GetFileDiff as the changed-code source of truth. Call it with exactly one prepared file: {"file_path":"<exact prepared path>"}
  • Use a returned cursor only for the same file. After invalid_arguments, correct the call once; do not repeat unchanged input.
  • Use Read, Grep, Glob, and LS only for context permitted by the prepared target evidence.
  • Never fetch, checkout, guess refs, run commands, or modify repository state.
  • Metadata hints orient the review but do not prove a finding. Verify every finding against the diff or permitted source context.
  • Preserve limited, stale, failed, omitted, conflicted, binary, or unavailable evidence as explicit coverage limitations. Missing evidence cannot become a clean result.

Primary review

For a strict run, inspect the target directly before considering delegation. For a prepared packet plan, inspect only enough manifest-level context to coordinate, then rely on packet-scoped workers and verify their findings without re-reading the whole large target:

  1. Understand the intended behavior and affected contracts.
  2. Trace changed paths far enough to confirm user-visible behavior, state transitions, errors, and compatibility.
  3. Check relevant trust boundaries, resource/concurrency behavior, module ownership, frontend behavior, and tests.
  4. Confirm each suspected issue before reporting it. Do not manufacture coverage by listing every possible domain.
  5. Record positive observations only when they are specific and useful.

Read the full file on GitHub · 75 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 · 75 lines · 0 tokens per session scan A 35f8e0e222e9

Subscribe to this mod's changes

deep_review_agent is an agent published in the GitHub repository GCWing/BitFun (1,871 stars, last pushed 2d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,128 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.