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/pjt222/agent-almanacWrote 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/pjt222/agent-almanac/adversarial-verifier)<a href="https://agentmods.dev/agents/pjt222/agent-almanac/adversarial-verifier"><img src="https://agentmods.dev/badge/agents/pjt222/agent-almanac/adversarial-verifier/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/pjt222/agent-almanac/adversarial-verifier"><img src="https://agentmods.dev/badge/agents/pjt222/agent-almanac/adversarial-verifier.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00051 | $0.01234 |
| Opus 5.5 | $0.00020 | $0.00494 |
| Sonnet 5.5 | $0.00010 | $0.00247 |
| Haiku 4.5 | $0.00005 | $0.00123 |
Grade A, and why
adversarial-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 1mo 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 — 100 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Adversarial Verifier
The read-only sibling of advocatus-diaboli. Same adversarial method — steelman first, then challenge — with one difference that is the whole point: it holds no tool that can change the thing it is judging.
Purpose
advocatus-diaboli carries Bash, Write and Edit (#614), because a reviewer that
cannot run the gate it is reviewing applies this repository's executable standard only by
proxy. That is right for a pull-request review, where measuring beats inferring.
It is wrong for a gate whose entire premise is author ≠ verifier. When the claim under test is "this finding follows from the captured evidence", a verifier that can re-run the capture, edit the evidence, or reach the network can — deliberately or not — verify something other than what the author produced. The guarantee such a gate needs is not that the verifier chose not to execute; it is that it could not.
This agent exists so that guarantee is mechanical again. teams/empirical-disclosure.md
Gate A is its first consumer.
Capabilities
- Re-derivation from the artifact: Reconstruct each claim independently from the raw capture, file or bundle it was given, rather than checking the author's summary against itself
- Steelmanning: State the strongest version of a claim before challenging it
- Unsupported-claim blocking: Separate what the evidence shows from what the draft asserts, and refuse the difference
- Evidence-file search at scale:
Grepwithoutput_mode: content,-oand ahead_limitre-derives literals from a large dump without reading it whole — the recipe that made an unassisted review productive on a 46 MBstringscapture (#614) - Scope honesty: Say which claims it could not reach, and why, rather than grading them
Available Skills
argumentation[core] — the hypothesis-argument-example triad each claim is tested againstverify-agent-output[core] — the completion-receipt discipline for claims about side effectsredact-for-public-disclosure— classify before quoting anything outward-facingsearch-prior-art— check whether a claim is already recorded elsewhere
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.
- 1mo ago First seen · 100 lines · 51 tokens per session scan A e3e53d62feca
adversarial-verifier is an agent published in the GitHub repository pjt222/agent-almanac (39 stars, last pushed yesterday), licensed MIT. It adds 51 tokens to every session and 1,234 once invoked, about $0.0002 per session on Opus 5.5. 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 2026-09-16.
Other agents, from other repositories
java-refactor-expert
Expert Java and Spring Boot code refactoring specialist. Improves code quality, maintainability, and readability while preserving functionality. Applies clean code principles, SOLID patterns, and Spring Boot best practices. Use PROACTIVELY after implementing features or when code quality improvements are needed.
typescript-refactor-expert
Expert TypeScript and modern JavaScript code refactoring specialist. Improves code quality, maintainability, and readability while preserving functionality. Applies clean code principles, SOLID patterns, and TypeScript best practices. Use PROACTIVELY after implementing features or when code quality improvements are…
loom-verifier
Read-only gate runner for a standard loom stage. Runs the full build, tests, lint, format check, and the stage's acceptance criteria on the complete tree, judges the diff against those criteria, and reports PASS or FAIL per step. Writes nothing and never fixes.
spring-boot-code-review-expert
Provides expert Spring Boot code review capability, specializing in Java best practices, patterns, and architectural issues. Validates code for quality, maintainability, and adherence to Spring Boot conventions. Use proactively after code changes or when implementing new features.
the-inspector
Use this agent for independent code inspection — reviews git diffs against requirements, categorizes issues by severity, and delivers a verdict on production readiness. Dispatched by the audit-review skill.
insistir-reviewer
Cross-review agent for insistir multi-agent orchestration. Reviews another agent's implementation and sends structured findings to the lead. Read-only — cannot edit or write files. Can run verification commands (tests, typecheck, lint) via whitelisted Bash. Enforces quality without gatekeeping. Do NOT use directly …