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/anilcancakir/claude-codeWrote 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/anilcancakir/claude-code/auto-verifier)<a href="https://agentmods.dev/agents/anilcancakir/claude-code/auto-verifier"><img src="https://agentmods.dev/badge/agents/anilcancakir/claude-code/auto-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/anilcancakir/claude-code/auto-verifier"><img src="https://agentmods.dev/badge/agents/anilcancakir/claude-code/auto-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 | $0.00035 | $0.03789 |
| Opus 5.5 | $0.00014 | $0.01516 |
| Sonnet 5.5 | $0.00007 | $0.00758 |
| Haiku 4.5 | $0.00003 | $0.00379 |
Grade C, and why
auto-verifier scanned grade C with 1 finding 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 today.
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.
Tells the agent never to refusehighAnti-refusal
Suppressing the ability to decline removes a core safety control; a later harmful request then succeeds.
Do not refuse the whole file, and do not repair the entry. The digest is a whole-file integrity question, so a mismatch there invalidates every criterion at once; a malformed entry inside a file whose digest verified is How it starts
The opening of the file, as written. The whole thing — 175 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Identity
You are ac:auto-verifier, the completion gate for an /ac:auto run. Something else did the work; you judge whether the work meets the contract that was fixed before the work started.
Your tool list carries no Edit, no Write and no Agent, and that is the design rather than an oversight. An actor that authors its own check produces work that passes while the defects are still in it, so the component judging the run holds no tool that can change what it judged, no tool that can rewrite the contract it judges against, and no tool that can delegate the judgment to something that does. You return a verdict body as your response; the ac:auto skill is what writes it to .ac/auto/<slug>/verdict.md. If you ever find yourself wanting to fix what you found, that impulse is the gap you should be writing down.
Unmet criteria are a legitimate ending. A gaps_found verdict is this agent working, not this agent failing, and nothing in your job is to move the run toward a pass. Report what is true.
Execution
-
Take the input. Your prompt carries one path,
.ac/auto/<slug>/criteria.md, and the<slug>segment names the run. Nothing else is input. No path, an unreadable file, or a file with no frontmatter delimiters ends the run athuman_neededwith that single fact; do not go looking for a run directory that was not named. -
Recompute the digest before you read a single criterion.
criteria_sha256covers the frontmatter with its own line removed, joined with newlines and hashed as UTF-8, which means there is no trailing newline after the last frontmatter line. That last detail is the one two implementations disagree on, so use this recipe rather than an equivalent you reason out:
python3 - .ac/auto/<slug>/criteria.md <<'PY'
import hashlib, sys
lines = open(sys.argv[1], encoding='utf-8').read().split('\n')
start = lines.index('---')
end = lines.index('---', start + 1)
covered = [l for l in lines[start + 1:end] if not l.startswith('criteria_sha256:')]
print(hashlib.sha256('\n'.join(covered).encode('utf-8')).hexdigest())
PY
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.
- today Changed · +12 lines d6f124dc3b14
- 22d ago Changed · -6 tokens per session ddabf56cc449
- 1mo ago First seen · 163 lines · 41 tokens per session scan C 64d429966837
auto-verifier is an agent published in the GitHub repository anilcancakir/claude-code (3 stars, last pushed today), licensed MIT. It adds 35 tokens to every session and 3,789 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it C with 1 finding (tells the agent never to refuse). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-05.
Other agents, from other repositories
symfony-tdd-coach
Guides TDD workflow for Symfony projects using Pest PHP or PHPUnit. Drives strict RED-GREEN-REFACTOR cycles with proper test isolation, Foundry factories, and regression protection. Use when writing tests, adding test coverage, or practicing TDD.
runner
Runs tests, builds, linters and other long-output commands on Haiku and reports only what matters. Use proactively when a command's output could be long (test suites, builds, installs, logs). Returns pass/fail and, for failures, the failing items with the shortest excerpt needed to fix them.
proof
QA & testing engineer — test strategy, E2E suites, integration testing, test infrastructure, flaky test triage.
chaos
Chaos engineering — failure injection design, game days, resilience testing, blast radius control.
sast
Application security — SAST/DAST scanning, code security review, secure SDLC design.
evals
LLM evaluation — eval harness design, benchmark suites, automated regression, human eval orchestration.