vet_skill_directory

Instructions

Implementation Reference

• src/openclaw_skill_vetter_mcp/analysis.py:137-153 (handler)

  Core handler function: vets every skill in a directory and aggregates results into a DirectoryVetReport. Calls vet_skill() for each skill, sorts by risk_score descending, and counts by risk level.

  def vet_skill_directory(skills: list[Skill], directory: str) -> DirectoryVetReport:
      """Vet every skill in a directory and aggregate."""
      reports = [vet_skill(s) for s in skills]
      reports.sort(key=lambda r: r.risk_score, reverse=True)
  
      by_level = Counter(r.risk_level for r in reports)
      return DirectoryVetReport(
          captured_at=datetime.now(UTC),
          directory=directory,
          skill_count=len(reports),
          skills=reports,
          aggregate_block_count=by_level[RiskLevel.BLOCK],
          aggregate_review_count=by_level[RiskLevel.REVIEW],
          aggregate_caution_count=by_level[RiskLevel.CAUTION],
          aggregate_clean_count=by_level[RiskLevel.CLEAN],
          aggregate_unknown_count=by_level[RiskLevel.UNKNOWN],
      )

• src/openclaw_skill_vetter_mcp/types.py:132-146 (schema)

  DirectoryVetReport model: the return type of vet_skill_directory, containing captured_at, directory, skill_count, per-skill VetReports, and aggregate counts by risk level.

  class DirectoryVetReport(BaseModel):
      """Aggregate report across all skills in a directory."""
  
      model_config = ConfigDict(frozen=True)
  
      captured_at: datetime
      directory: str
      skill_count: int
      skills: list[VetReport]
      """Per-skill reports, sorted by risk_score descending."""
      aggregate_block_count: int
      aggregate_review_count: int
      aggregate_caution_count: int
      aggregate_clean_count: int
      aggregate_unknown_count: int

• src/openclaw_skill_vetter_mcp/server.py:67-75 (registration)

  Tool registration in list_tools(): defines the 'vet_skill_directory' Tool with description and empty inputSchema (no required arguments).

  Tool(
      name="vet_skill_directory",
      description=(
          "Run all scanners on every skill in the configured directory and return "
          "an aggregate report (per-skill VetReports + counts by risk level). "
          "Use this for a periodic audit of installed skills."
      ),
      inputSchema={"type": "object", "properties": {}, "required": []},
  ),

• src/openclaw_skill_vetter_mcp/server.py:147-150 (registration)

  Tool dispatch in call_tool(): when name=='vet_skill_directory', fetches skills and directory from backend, calls vet_skill_directory(), and serializes the result.

  if name == "vet_skill_directory":
      skills = await backend.get_skills()
      directory = await backend.get_directory()
      return _serialize(vet_skill_directory(skills, directory))

• src/openclaw_skill_vetter_mcp/server.py:303-305 (helper)

  Helper _serialize(): converts any Pydantic model to MCP TextContent (JSON-indented) for the response.

  def _serialize(model: Any) -> list[TextContent]:
      """Pydantic model → MCP TextContent (single block, JSON-serialized)."""
      return [TextContent(type="text", text=model.model_dump_json(indent=2))]