Openclaw Skills turns AI product details into the right CAC filing path, required materials, submission steps, and ongoing compliance actions.
The fastest way to install a skill directly from the registry.
npx clawhub@latest install algorithm-filing
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
Copy this prompt to OpenClaw to install it automatically.
Help me install algorithm-filing using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Get the raw skill files in a ZIP archive.
Openclaw Skills is a regulatory filing assistant for AI large models and algorithms under China CAC requirements. It analyzes your product scenario, determines whether you need a large model filing, large model registration, algorithm filing, dual-new assessment, or no filing, then converts that decision into an actionable compliance plan.
It is built for teams shipping public-facing AI services, wrapped third-party model APIs, personalization/recommendation systems, and other algorithm-driven products. Openclaw Skills helps you prepare the filing materials, generate a security assessment report framework, guide system submission on beian.cac.gov.cn, and maintain post-filing compliance across changes, annual checks, logging, and content review.
# No CLI installation is required.
# Provide the agent with a product summary such as:
# "I have a self-developed text generation model serving the public via API."
# Then request:
# "Determine the filing type, required materials, and security assessment framework."
| Layer | Artifact | Purpose | Source Rules / Taxonomy |
|---|---|---|---|
| Input profile | Product and organization description | Captures the facts needed to determine compliance scope | entity type, public-facing status, model origin, API/app wrapping, personalization, ranking/filtering, output media, internal-only use |
| Filing classification | Compliance decision matrix | Maps the scenario to the correct filing path | large model filing, large model registration, algorithm filing (generative synthesis, other), dual-new assessment, no filing, change filing, annual review, cancellation |
| Materials package | Material checklist | Produces the required submission attachments | business license, legal rep ID, responsible person ID/work proof, ICP filing/permit, commitment letter, algorithm responsibility statement, product-linked evidence |
| Report framework | Security assessment outline | Structures the filing report by required dimensions | large model: data safety, model safety, generated content safety, protective measures; algorithm: template-based multi-dimension risk assessment and defenses |
| Filing workflow | Submission playbook | Guides form completion and review steps | subject registration, algorithm info submission, product-function binding, approval waiting, revision after rejection |
| Post-filing compliance | Ongoing control plan | Tracks obligations after approval | public notice, change reporting within 10 working days, termination cancellation within 20 working days, annual self-checks, log retention, content review |
| Rejection remediation | Correction log | Converts review feedback into a revised submission | rejection reason, impacted section, evidence gap, fix plan, resubmission status |
| Reference files | Markdown knowledge assets | Standardizes reusable guidance | references/material-checklist.md, references/security-assessment-framework.md, references/filing-workflow.md, references/common-rejection-reasons.md |
filing_type: large_model_filing, large_model_registration, algorithm_filing_generative, algorithm_filing_other, dual_new_assessment, no_filingstage: planning, material_prep, document_drafting, system_submission, review, post_filing, remediationrisk_area: data_legality, content_filtering, copyright, personal_information, training_safety, prompt_injection, minors_protection, public_opinioncontrol_metric: external_data_share, risk_coverage, content_test_count, refusal_test_count, manual_review_rate, log_retention_months, compliance_threshold| Metric | Target |
|---|---|
| External data share | <= 30% |
| Risk coverage | >= 17 categories |
| Content test questions | >= 2000 |
| Refusal test questions | >= 500 |
| Sensitive-scene refusal rate | >= 95% |
| Manual review rate | >= 10% |
| Log retention | >= 6 months |
| Security assessment report length | 60-90 pages |
| Personal information detection after masking | 0 |
| Model safety test coverage | 31 risk scenarios |
| Overall compliance threshold | >= 90% |
Loading
Openclaw Skills for Taiwan property valuation converts comparable sales and adjustment factors into defensible market prices.

An Openclaw Skills home-repair diagnostic assistant that turns vague faults into safe, step-by-step fixes, tool lists, and repair-vs-pro advice.

Openclaw Skills for migrating OpenAI-compatible apps to TokenLab with minimal code changes, live model discovery, and preserved request semantics.

Openclaw Skills for Kubernetes triage automation detect failures, correlate logs and events, and produce root-cause reports in seconds.

A highly efficient utility to convert raw log files into Prometheus metrics, JSON, or CSV formats using customizable regex patterns.

An MCP-compliant API providing structured data for over 260 Hong Kong insurance products, including internal rates of return (IRR) and historical dividend fulfillment statistics.








































