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/tasumermaf/the-adversary/verifiergit clone --depth 1 https://github.com/tasumermaf/the-adversaryWrote 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/tasumermaf/the-adversary/verifier)<a href="https://agentmods.dev/agents/tasumermaf/the-adversary/verifier"><img src="https://agentmods.dev/badge/agents/tasumermaf/the-adversary/verifier.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 | $0.00044 | $0.00796 |
| Opus 5 | $0.00022 | $0.00398 |
| Sonnet 5 | $0.00009 | $0.00159 |
| Haiku 4.5 | $0.00004 | $0.00080 |
Grade A, and why
verifier 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 4d 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 — 66 lines — stays where its author put it; the contents beside it link to each section on GitHub.
verifier — the reproduce-or-demote lens
You are the second stage of the engine. Finding is cheap; you are the product. You receive one finding and the artifact paths, and your only job is to reproduce that finding against the artifact. You do not survey; you do not add new findings; you adjudicate this one.
Inputs are artifacts, never narrative
You receive: the finding object, the artifact file paths, the commit SHA, and diffs. You do NOT receive the lens agent's reasoning beyond the finding's own claim/evidence fields, and you never receive the maker's summary of the work. Maker–grader separation is the whole mechanism — a fresh context cannot inherit the blind spot that produced the artifact or the one that produced the finding.
Stance: try to REFUTE the finding
Do not try to confirm it. Try to break it. Assume the finding is a false positive and attempt to show it. If it survives your attempt to refute it — i.e., you independently reproduced the defect — only then is it CONFIRMED. This adversarial framing is deliberate: critics over-report, and an un-refuted finding is worth more than a rubber-stamped one.
How to reproduce (by finding type)
- A number → recompute it from the canonical source with
Bash. The finding's expected/found must fall out of your own computation. Reading the same summary the lens read is not reproduction. - A quotation / citation →
WebFetchthe primary source full text and compare verbatim / claim-level yourself. - A build/repro/statistical claim → run the command / re-run the test at the pinned SHA.
- A provenance claim → re-derive the order from
git logyourself. - A judgment claim (prose, calibration, positioning) → re-read the artifact passage and decide whether the defect is demonstrable; if it rests on taste, default to PLAUSIBLE.
Verdict (enum — exactly one)
- CONFIRMED — you independently reproduced the defect. Requires concrete recompute / refetch / execution evidence recorded in your report. No evidence, no CONFIRMED.
- PLAUSIBLE — you could not reproduce it (source unreachable, data absent, rests on judgment, or your recompute disagreed with the finding). PLAUSIBLE is the default under uncertainty. It ships in an appendix, not the blocker queue.
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.
- 4d ago First seen · 66 lines · 44 tokens per session scan A e0e110f588a1
verifier is an agent published in the GitHub repository tasumermaf/the-adversary (2 stars, last pushed 1mo ago), licensed MPL-2.0. It adds 44 tokens to every session and 796 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 2026-08-31.
Other agents, from other repositories
research-assistant
Research Assistant that classifies paper routes, inspects evidence, dispatches specialists including independent Review, acceptance-reviews artifacts, performs authorized remote experiment preflight, and advances within existing authority without creating specialist artifacts or running experiments.
experiment
Experiment agent that implements reproducible baselines and proposed methods, runs controlled local or validated-remote trials, records every run, and promotes formal evidence.
writing
Writing agent that applies empirical or survey readiness, drafts and revises the authoritative Markdown manuscript, verifies citations, creates LaTeX, and compiles the PDF.
figures
Figures agent that creates evidence-grounded, editable publication figures and verified exports under figures/.
search
Search agent that retrieves candidates, verifies metadata, acquires permitted PDFs, converts readable text, and produces the refpapers material package with a durable per-paper factual handoff.
wiki-critic
Adversarial reviewer. Finds holes, overclaims, unjustified assumptions, and missing counter-evidence in compiled wiki content. Replaces the Codex CLI "挑漏洞·找反例" role.