read-prior — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited read-prior (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Validate claims from check_claim.json that require external evidence.
{paper_dir}/reports/check_claim.json- [ ] Step 1: Load target claims
- [ ] Step 2: Search and judge each claim
- [ ] Step 3: Save reportRead {paper_dir}/reports/check_claim.json. Collect all claims where evidence_type includes "related_work".
Each claim's id, source, quote, claim, evidence_type fields should be copied as-is into the output — do not modify them.
These cover three sub-types — use the quote and claim text to identify which:
For each claim, run WebSearch queries. When a relevant paper is found, fetch its full page (arXiv abstract or PDF landing page) and look for specific supporting details — e.g., which table reports the result, which section describes the method, what numbers are shown. Include these details in reason (e.g., "Table 2 in [paper] reports X% on [benchmark] under the same setting").
Assign a status:
| Status | Meaning |
|---|---|
pass | Web evidence confirms the claim |
warning | Claim is imprecise, overstated, or has a missing citation |
error | Claim is factually wrong or clear prior art exists |
unverifiable | Insufficient web evidence |
For novelty claims, also note if the novelty is novel, incremental, or prior_art_exists.
When status is warning or error, include a suggest field with a one-sentence concrete fix (e.g., "Relabel as incremental and cite Chen et al. 2021").
Output: {paper_dir}/reports/check_prior.json
{
"summary": {
"total": 5,
"pass": 3,
"warning": 1,
"error": 0,
"unverifiable": 1
},
"results": [
{
"id": 12,
"source": "introduction",
"quote": "We are the first to apply mutual information maximization to knowledge distillation",
"claim": "No prior work has used mutual information for knowledge distillation",
"evidence_type": ["related_work"],
"status": "warning",
"reason": "Chen et al. (2021) proposed MI-based distillation for NLP; this paper's novelty is limited to the vision domain",
"suggest": "Relabel the novelty as `incremental`, restrict the claim to the vision domain, and cite Chen et al. (2021)."
}
]
}Every claim with `related_work` in evidence_type MUST appear in `results`. No claim may be silently skipped. If a claim's cited fact checks out quickly, still include it with status: "pass" and the source.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.