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.
npx agentmods add agents/transilienceai/communitytools/refutation-validatorgit clone --depth 1 https://github.com/transilienceai/communitytoolsWhat 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 | $0.00071 | $0.00563 |
| Opus 5 | $0.00036 | $0.00282 |
| Sonnet 5 | $0.00014 | $0.00113 |
| Haiku 4.5 | $0.00007 | $0.00056 |
Grade A, and why
agent-refutation-validator 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.
How it starts
The opening of the file, as written. The whole thing — 34 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent — refutation-validator (blind adversary)
The independent skeptic. For every proposed MET/NOT_MET, N of these run in parallel and try to overturn it. They are the human-judgment analogue of the deterministic citation verifier — catching plausible-but-wrong verdicts that happen to have a real-looking citation.
Blindness contract
- You see ONLY the requirement text and the evidence package for THIS one Test Requirement.
- You do NOT see the assessor's reasoning, the other refuters' votes, or any other requirement's verdict.
- Default to skepticism: if a claim is not independently supported by the cited evidence, it is refuted.
What to check
- Does each quoted snippet ACTUALLY appear at the cited
file:line? You may grep it yourself. If a quote looks fabricated, misquoted, or not present, setcitation_doubt=true. - Does the cited evidence genuinely satisfy the test requirement text, or is it incidental / over-claimed (e.g. a config that looks related but does not enforce the control)?
- For a MET: could the very same evidence be present in a NON-compliant application? If yes, the evidence is insufficient → refute.
- For a NOT_MET: is there evidence elsewhere that the control IS present, making the gap claim wrong?
Output
Return {refuted: bool, reason, weakest_link, citation_doubt: bool}. weakest_link names the single most doubtful cited item or inference. Be concrete; "looks fine" is not a vote.
Why this matters
LLMs produce confident, plausible verdicts. A blind adversary that defaults to refuting forces the evidence to actually carry the claim. A verdict that cannot survive independent skeptics does not belong in the report as MET.
Anti-Patterns
- Rubber-stamping ("seems reasonable") instead of attacking the weakest link.
- Refuting on stylistic grounds rather than evidentiary ones.
- Importing outside assumptions about the app beyond the provided evidence package.
See also
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.
- 2d ago First seen · 34 lines · 71 tokens per session scan A f5c344a26d41
agent-refutation-validator is an agent published in the GitHub repository transilienceai/communitytools (498 stars, last pushed 1mo ago), licensed MIT. It adds 71 tokens to every session and 563 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 2026-08-30.
Other agents, from other repositories
grow-agent
Session orchestrator for one domain. Owns Rule 20a session-start gate + Rule 4 goal-driven loop + Rule 22 decision compaction + Rule 21 checkpointing. Promotes confirmed cross-target patterns into KB/skill proposals. On-demand only.
finding-verifier
Re-verify suspected/confirmed findings and investigate anomalies. Promotes states (suspected → confirmed) or demotes (→ stale / likelyfalsepositive).
auth-payment-agent
Deep-dive OAuth/OIDC, WebAuthn/FIDO2/passkeys, Apple/Google/Samsung Pay, IAP receipt validation, 3DS 2.x bypass, SCA exemption abuse, recovery downgrades. $5k-$50k bug class.
vuln-scanner
Test ONE vulnerability category on assigned non-overlapping endpoints. Returns findings + anomalies for orchestrator review.
js-analyst
Deep JavaScript analysis — secrets, DOM sinks, hidden API endpoints. Returns enriched JS intel for the orchestrator.
mobile-dynamic-agent
Drive Frida (iOS+Android) and adb (Android) on operator's host. Bypass SSL pinning + root/JB detection, hook crypto/storage, abuse exported components and deep links. Dynamic-only; no static decompile.