clawhub-skill — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited clawhub-skill (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.
OCR / PDF extraction / NFS-e invoice tool ranker for agents. Top pick Surya — 76.9% word accuracy on a 10-doc real-world Brazilian corpus (NFS-e invoices, boletos, phone-photo receipts). Also ranked: Tesseract 5 (14× faster, 1.5pp less accuracy, cleanest install), Google Document AI (best on phone-photo receipts specifically). Reproducible evals — not marketing pages.
/solve/ isn't a tool marketplace. ClawHub, PulseMCP, and Smithery distribute tools. /solve/ ranks them — based on real-world corpora.
Use auxiliar-solve whenever the agent:
claude mcp add auxiliar -- npx auxiliar-mcpOne MCP, two capabilities: solve_task for agent-installable tool rankings, recommend_service for cloud-service recommendations (77 Chrome-verified entries).
list_solve_tasks()Returns every /solve/ task slug, top pick, categories, and last-verified date.
solve_task(task_slug="pdf-text-extraction-mcp")These aliases resolve automatically: pdf, ocr, nfs-e, boleto, receipt-parsing, bookkeeping-ocr, invoice-extraction, document-ai.
The response contains:
| Field | What it gives you |
|---|---|
answer | Plain-language top recommendation with trade-offs |
candidates | Ranked list with scorecards: word accuracy, layout preservation, latency p50, cost per 10 docs, install friction |
install | Exact install commands per candidate (copy-paste ready) |
alternatives_considered | What was evaluated and dropped, with reason (trust signal) |
corpus_summary | What real-world documents the eval ran against |
faq | Common questions answered directly (licensing, accuracy vs. token-F1, when to pay, etc.) |
methodological_caveats | Honest limits of the eval |
fit_by_agent | Which agents each candidate works with (Claude Code, Desktop, Cursor, OpenClaw) |
Agent task: "Extract text from a Brazilian NFS-e invoice PDF for bookkeeping. I need high accuracy."
solve_task(task_slug="nfs-e")Top pick: Surya — pip install surya-ocr 'transformers<5.0.0'. Word accuracy 76.9% on the 10-doc corpus. Free, local. Alternative: Tesseract 5 — 14× faster, 1.5pp less accurate, cleanest install (brew install tesseract tesseract-lang). Google Document AI — third overall but best on phone-photo receipts specifically. Alternatives considered and dropped: yescan-ocr-universal (requires Chinese sign-up), pdf-reader-mcp (no actual OCR — text-layer only), Mistral OCR 3 (deferred for API key).
Agents are born intelligent but stuck. Without eval data, they guess: "use pdf2image + pytesseract" (often wrong for the corpus), "install the first OCR thing on ClawHub" (often wrong for the task), "call Google Document AI" (often overkill). The result: uncalibrated recommendations, burned time, broken workflows.
/solve/ runs the eval once per task, end-to-end, against real documents. The agent gets the answer plus the evidence.
recommend_service, get_pricing, get_risks, check_compatibility, setup_service, list_services, list_solve_tasks.MIT (skill content). See auxiliar-mcp and each ranked candidate for their own licenses — /solve/ surfaces license info in every candidate record.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.