crw-research — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited crw-research (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.
Find EVERY arXiv paper that answers a research query. Recall = union of arXiv ids; extra ids never hurt, so cast wide but on-topic. The Research API is a live, drop-in Firecrawl-research-compatible surface — your job is query strategy + intent routing, the endpoints do the retrieval.
Set FASTCRW_API_KEY (a crw_live_… key from https://fastcrw.com/dashboard). Base URL https://api.fastcrw.com. Every endpoint is a GET; pull arXiv ids out of results[].ids.arxiv / results[].primaryId.
# search: ranked papers for one query
curl -s -H "Authorization: Bearer $FASTCRW_API_KEY" \
"https://api.fastcrw.com/v2/search/research/papers?query=$(jq -rn --arg q "QUERY" '$q|@uri')&k=40"
# references / citers / similar of a seed paper (citation graph)
curl -s -H "Authorization: Bearer $FASTCRW_API_KEY" \
"https://api.fastcrw.com/v2/search/research/papers/arxiv:1706.03762/similar?intent=related%20work&mode=references&k=40"A) ALWAYS (base): write 8–12 exact-name queries — specific method, model, dataset, and benchmark NAMES, not broad phrases ("MoleculeNet benchmark", "Uni-Mol", "ChemBERTa", not "molecular embeddings"). Call search on each, union the arXiv ids, rank by how many queries surfaced each id. Exact-name decomposition is the #1 recall lever — one broad query misses the niche papers.
B) COMPARE-AGAINST ("what does X compare to / build on / baseline against") → resolve X to its arXiv id, then /papers/arxiv:<X>/similar?mode=references. The answer lives in X's own bibliography.
C) USING / EXTENDING X ("models that USE/adopt X") → /similar?mode=citers (forward citations) + exact-name searches for known adopters.
D) BEST-ON-BENCHMARK ("which models score best on X", "largest open model") → search the leaderboard, read the OPEN model names (DeepSeek/Qwen/GLM/Kimi/MiniMax/Llama/Mistral/Gemma — ignore Claude/GPT/Gemini, no papers), then search "<model family> technical report" for each.
E) NICHE ENUMERATION ("papers that do X") → exact-name queries (A) are primary. A tight survey or awesome-list, when on-topic, adds its ids.
the family, don't stop at one. Only a query naming a paper by title is single.
exact-name hits first, broad-search tail after. Never invent ids.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.