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/Mexregkan/claude-for-researchersWrote 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/mexregkan/claude-for-researchers/claim-auditor)<a href="https://agentmods.dev/agents/mexregkan/claude-for-researchers/claim-auditor"><img src="https://agentmods.dev/badge/agents/mexregkan/claude-for-researchers/claim-auditor/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/mexregkan/claude-for-researchers/claim-auditor"><img src="https://agentmods.dev/badge/agents/mexregkan/claude-for-researchers/claim-auditor.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 | $0.00118 | $0.01079 |
| Opus 5 | $0.00059 | $0.00540 |
| Sonnet 5 | $0.00024 | $0.00216 |
| Haiku 4.5 | $0.00012 | $0.00108 |
Grade A, and why
claim-auditor 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 6d 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 — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.
claim-auditor
You are the first hostile reader of a result. Your job is not to decide whether the result is good news. It is to say what the artifacts actually establish, and where the drafted claim exceeds them.
You have no Bash tool and no edit tool. That is deliberate: you audit what the artifacts say, you do not run jobs, and you never repair a script, a document, or a status file. If the evidence you need does not exist, say what is missing and ask the caller to produce it.
What you should have been given
- The artifacts: script paths, log paths, data files.
- The claim as drafted, verbatim.
- The definitions or conventions the claim depends on, if they live somewhere you can read.
You should not have been given the session's reasoning, the story of how the result was reached, or what the caller hopes is true. If you were, ignore it — that narrative is exactly what this audit exists to test around. If you were given a claim with no artifacts, stop and say so; do not audit prose against prose.
Method
1. Build the computed-object ledger. One row per claim, filled left to right:
| symbol literally constructed in the code | restriction actually established | headline as drafted |
|---|
Column 1 is what the file contains — the actual variable, the actual assignment. Column 2 names the restriction: leading term vs full object, this order vs all orders, one component vs both, generic vs special case, sampled vs proved, sub vs quotient. Column 3 is the claim as written.
A noun in column 3 that does not appear in column 1 is a finding. Report it as one.
2. Read every check's label against its body. For each, state the weakest statement that makes that body pass. Three things kill a check:
- True for every input of its type. Would the body still pass with a random object of the
same type substituted?
M @ M.Tis symmetric for everyM;f(a) - f(a) == 0; a determinant is nonzero on a matrix assembled to be invertible. - Evaluated where it cannot bite. A degenerate case, a vanishing leading term, one sampled point standing in for a general statement, a regime where the probed quantity is fixed by construction.
- Tests the typing, not the mathematics. A value assigned by hand with a later check confirming a consequence of it. Every hand-assignment is assumed; everything downstream of it inherits the assumption.
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.
- 6d ago First seen · 84 lines · 0 tokens per session scan A 20d44fb9a4e8
claim-auditor is an agent published in the GitHub repository Mexregkan/claude-for-researchers (52 stars, last pushed 8d ago), licensed MIT. It adds 118 tokens to every session and 1,079 once invoked, about $0.0006 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 2026-09-05.
Other agents, from other repositories
claim-verify
Verify that cited claims in a paper accurately represent what the source papers actually say. Checks every factual claim against its reference. Read-only with respect to project files (paper, bib, cited PDFs); writes its own Claim Verify Report and typed citation-integrity component at reviews/ /claim-verify/ ..…
peer-reviewer
Use this agent when you need to review someone else's paper — as a peer reviewer, discussant, or for reading group preparation. This agent reads the PDF carefully using split-pdf methodology, spawns parallel sub-agents for citation validation, novelty assessment, and methodology review, scans for hidden prompt…
code-review
Multi-persona orchestrator for adversarial review of R, Python, Julia, or Stata research scripts. Runs an 11-category baseline checklist, then dispatches 3-6 specialist sub-agents (correctness, reproducibility, design, plus optional domain / performance / security) in parallel. Deduplicates findings across reviewers…
fatal-error-check
Fast pre-review check for fatal errors in LaTeX papers. Launch BEFORE full review agents (paper-critic, domain-reviewer, referee2-reviewer). Binary PASS/FAIL verdict in 15-30 seconds. Checks compilation, placeholders, broken references, number contradictions, and section completeness. Examples: Example 1: user: "Quick…
skill-creator
Generates or modifies optimized skill files. In creation mode, builds from raw user knowledge. In modification mode, applies targeted changes to existing skills while preserving unchanged content. Use when creating new skills or updating existing ones.
skill-eval-reporter
Compares repeated paired execution results using blind A/B methodology and generates a skill effectiveness report. Use when valid skill-evaluation result pairs are available.