The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/match_skills.py executes the local scan_registry.py script using subprocess.run. This is used to ensure the skill registry is synchronized before performing capability matching. The execution is limited to a known local script path and uses the current Python interpreter.
[DATA_EXPOSURE]: scripts/scan_registry.py performs a filesystem walk to identify SKILL.md files. It extracts metadata including name, description, and tags to build a local registry. This access is restricted to the skill's root and standard agent skill paths.
[PROMPT_INJECTION]: The skill is designed to process external skill metadata which represents a surface for indirect prompt injection. A malicious skill file could theoretically use its description field to influence the orchestrator's ranking logic. However, the orchestrator uses yaml.safe_load() in scripts/scan_registry.py to prevent unsafe deserialization during metadata extraction.
[SAFE]: No remote code execution, external data exfiltration, or persistence mechanisms were detected. The skill's behavior is consistent with its stated purpose of agent orchestration.

agent-orchestrator