Skip to content

Agent Skills

Agent skills are SKILL.md files that teach a coding agent how to use WhiteBoxXAI correctly: which SDK calls and MCP tools exist, what their arguments mean, and what the agent must never do (invent a score, commit a key, cast a governance vote). Each skill has a TRIGGER / SKIP description so the agent loads it only when it applies.

All skills are MIT-licensed, published by AgentaFlow, and checked against the current SDK and MCP tool list before release.

Install

Pick one:

Read https://whiteboxxai.com/.well-known/skills/whitebox-getting-started/SKILL.md and set up WhiteBoxXAI for this repository.

npx skills add https://whiteboxxai.com

Lists every skill and installs the ones you pick.

Download whitebox-skills.zip and unzip it into your agent's skills directory, for example .claude/skills/ or .agents/skills/.

Harnesses that support skill discovery can read the index at https://whiteboxxai.com/.well-known/skills/index.json. Each entry carries a SHA-256 digest so the harness can verify the file it downloads.

The skills

Always load whitebox-governance-rules alongside any other WhiteBoxXAI skill.

Skill Load it when the user wants to... Skip it when...
whitebox-getting-started Add or connect WhiteBoxXAI: install the SDK or MCP server, set an API key, choose an integration path WhiteBoxXAI is already configured
whitebox-governance-rules Do anything that reads from or writes to WhiteBoxXAI The work never touches WhiteBoxXAI
whitebox-register-and-monitor-models Register and log predictions for scikit-learn, XGBoost, LightGBM, PyTorch, TensorFlow/Keras or Hugging Face models The system is an LLM, RAG or agent app
whitebox-drift-alerts-explainability Check drift, set alert rules, explain predictions with SHAP/LIME, run a fairness audit The model isn't registered yet
whitebox-llm-and-agent-observability Log LLM calls, RAG retrievals, safety checks and multi-agent workflows; version, deploy and roll back prompts; group a run into one Governance Event The system is a tabular ML model
whitebox-mcp-server Connect an assistant, IDE or agent harness over MCP, or pick the right MCP tool Instrumenting application code (use the SDK skills)
whitebox-compliance-evidence Summarise ISO/IEC 42001 readiness, the risk register, review dossiers, or reports for a risk officer or auditor The request needs a legal conclusion

The full index with descriptions is at whiteboxxai.com/skills/index.md.

What the skills enforce

The skills say more than "how to call the API". They also carry the rules a compliance team depends on:

  • Report empty, errored or unassessed results as they are, never as a default value.
  • Put a plain-language verdict beside every drift statistic, fairness metric or explanation weight.
  • Read keys from the environment; never write one into source, notebooks or logs.
  • Treat a 404 on an id as "not in this organization", not as an outage.
  • Never cast a governance vote on a person's behalf.
  • Stay within ISO/IEC 42001, GDPR, CCPA, the EU AI Act and NIST AI RMF.

See Agent & Developer Resources for the other agent entry points.