mantis-review

A review step for checking reported software findings against the actual source code and marking whether they are valid or false alarms.

In plain words
What is it for?
Use it to validate findings, record the reasoning, add reproduction hints, and update their review status.
Why use it?
It helps separate real problems from reports that do not hold up when the code is inspected.

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/google/mantis/mantis-review
Any agent
npx skills add google/mantis --skill mantis-review
Clone the repo
git clone --depth 1 https://github.com/google/mantis

Made for: Claude Code, Codex.

Per session 38 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 6,345 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 $0.00038 $0.06345
Opus 5 $0.00019 $0.03173
Sonnet 5 $0.00008 $0.01269
Haiku 4.5 $0.00004 $0.00634

Measured 2d ago against content hash 3c0a274abc31, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

mantis-review 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.

mantis-review/SKILL.md · 438 lines

How it starts

The opening of the file, as written. The whole thing — 438 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Reviewer (/mantis-review)

System Goal

Independent Validator. Reviews consolidated findings against active source code to verify validity and filter out noise and false positives.

Command Definition

  • Command: /mantis-review
  • Description: Independently reviews findings and filters out false positives.

Input/Output Contract

  • Reads:
    • workspace/findings/ (finding JSON files, including each finding's optional discovery_commit provenance stamp).
    • workspace/.mantis_state.json (current loop pass_number; and, when present, active_snapshot = {root, snapshot_id, snapshot_pinned} for snapshot provenance).
    • Invocation args when passed by the orchestrator: --snapshot_root, --snapshot_id, --state_root (and, rarely, --target_root).
    • Target source code files at the code_paths paths/lines, read under the resolved CODE_ROOT (the pinned snapshot root — see Instructions Step 0 / Block A), never the live tree when a snapshot is pinned.
  • Writes:
    • Updates findings on disk in-place (sets "status", "reasoning", "repro_hints", "triage_checklist", and appends history).
    • Writes helper script workspace/helpers/append_review.py.
  • Preconditions:
    • workspace/findings/ exists with finding files.
    • Target source code files exist.
  • Idempotency Guarantee:
    • Modifies finding files in-place using the helper script append_review.py, which MUST carry # MANTIS_HELPER_VERSION = 2 as its first line; regenerate the helper if that marker is absent or a different integer (see Instructions step 5).
    • Idempotency key is (pass_number, snapshot). Before (re)writing a finding, scan its history array:
      • SNAPSHOT MODE OFF (neither --snapshot_id was passed NOR is active_snapshot readable in state): SKIP the review update iff a "stage": "reviewer" entry already exists for the current pass_number (byte-for-byte today's behavior).
      • SNAPSHOT MODE ON: SKIP the review update iff a "stage": "reviewer" entry exists whose pass_number equals the current pass AND whose snapshot equals SNAPSHOT_ID. A reviewer entry that lacks snapshot (legacy) or carries a different snapshot counts as UNKNOWN → RE-REVIEW (do NOT skip), so a finding reviewed under the old skill is re-grounded against the current pinned snapshot.

Read the full file on GitHub · 438 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 · 438 lines · 38 tokens per session scan A 3c0a274abc31

Subscribe to this mod's changes

mantis-review is a skill published in the GitHub repository google/mantis (853 stars, last pushed 5d ago), licensed Apache-2.0. It adds 38 tokens to every session and 6,345 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 2026-08-30.