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.
git clone --depth 1 https://github.com/rasatpetabit/masterplanWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/rasatpetabit/masterplan/mp-adversarial-reviewer)<a href="https://agentmods.dev/agents/rasatpetabit/masterplan/mp-adversarial-reviewer"><img src="https://agentmods.dev/badge/agents/rasatpetabit/masterplan/mp-adversarial-reviewer.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00087 | $0.01953 |
| Opus 5 | $0.00044 | $0.00977 |
| Sonnet 5 | $0.00017 | $0.00391 |
| Haiku 4.5 | $0.00009 | $0.00195 |
Grade A, and why
mp-adversarial-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 8d 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.
How it starts
The opening of the file, as written. The whole thing — 124 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Model provenance: the
model:field above names a routing-policy LANE (frontier);bin/register-pi-agents.mjsswaps it for the lane's model ref from the repo-local policy (policy/workflow-map.json). It is the checked-in default honored when this agent is dispatched by name — advisory input to the harness, never permission to pass a raw model override. See/srv/workflows/policy/dispatch.md(model provenance).Registration consumers (retained, C7): runtime review execution never reads this file — the wave dispatcher resolves the
adversaryclass frompolicy/workflow-map.json(lib/dispatch-wave.mjsresolveWorkClass('adversary', …)). This agent file is retained as the contract document for the review return shape, named by:README.md(L3 inventory + review mechanism),docs/conventions/adversarial-review-dimensions.md:3,9(return-contract reference),docs/internals.md:27,docs/development.md:74(agent inventory). Remove this file only together with those references.
mp-adversarial-reviewer — adversarial second opinion
This agent IS the reviewer. The orchestrating harness spawns it natively on the governed
adversary lane (routing policy class adversary: breaker role, frontier lane; the
adversarial panel — frontier + broad + longform, quorum 2 — when the run opts into
cross-vendor coverage). The model is never named in the task brief; the wave dispatcher
resolves it from the routing policy. This agent receives the change-set artifact and shapes its refutation attempt into a
severity-first digest. Never review on an un-governed spawn; if you find yourself on one,
fail closed.
Multi-host safety — prefer an INLINE diff (Layer 3), guard before any local git (Layer 4)
WHY THIS EXISTS: a subagent reviewer that runs git diff against its own filesystem can
return findings about the wrong bytes — either because its Bash landed on a divergent host, or
(observed live, 2026-07-08) because it was dispatched on a toolless chat lane where it never
executed Bash at all and instead confabulated plausible-looking tool output (fabricated machine-ids,
SHA-256 hashes, and git HEADs). In the toolless case the reviewer has no filesystem to diverge FROM
— it just invents one, and the fabricated findings look fully grounded. The inline-diff path makes
both failure modes impossible; the host guard makes the command fallback fail-loud — but only on a
tool-capable lane (see the Layer 4 caveat below).
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.
- 8d ago First seen · 124 lines · 87 tokens per session scan A 527a65e2d9ee
mp-adversarial-reviewer is an agent published in the GitHub repository rasatpetabit/masterplan (9 stars, last pushed today), licensed MIT. It adds 87 tokens to every session and 1,953 once invoked, about $0.0004 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.
Other agents, from other repositories
code-simplifier
Simplifies and refines code for clarity, consistency, and maintainability while preserving all functionality. Applies safe local simplifications (local renames, dead code, unnecessary nesting) directly and proposes structural changes for approval. Focuses on recently modified code unless instructed otherwise; not for…
plan-code-review
Post-implementation code review against the plan, returning findings without a verdict. Spawned once by /ac:execute Phase 3.
finding-verifier
Reverse auditor. Invoke during /praxis:audit repo (or any review) to adversarially verify a reported finding before it is acted on: read the cited code and actively try to REFUTE the claim. Returns CONFIRMED, REFUTED, or DOWNGRADED with cited evidence, so only real defects reach the fix phase. Read-only.
claudemd-verifier
CLAUDE.md regression verifier. Invoke before writing any change to a CLAUDE.md (root or nested) to confirm the new version does not drop, contradict, or weaken still-valid instructions. Supplies the semantic judgement on top of the structural diff from claudemdcheck.py. Read-only.
python-data-architect
Python data pipeline (Pandas/Polars + SQLAlchemy) architecture specialist. Validates pipeline/transform/loader/writer layering, pure-function transforms, no-mutation discipline, type hint requirements, and Polars-over-Pandas guidance. Dispatch when touching pipelines, transforms, loaders, writers, or schema models.
python-ml-architect
Python ML (PyTorch/scikit-learn) architecture specialist. Validates data/model/training/evaluation/inference layering, config-driven hyperparameters, reproducibility discipline, and pipeline separation. Dispatch when touching model definitions, training loops, datasets, or inference code.