skill-registry

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill performs broad scans of sensitive global directories in the user's home path, including ~/.claude/skills/, ~/.cursor/skills/, ~/.gemini/skills/, and ~/.config/opencode/skills/. It extracts content from these files and persists it into the local project's .atl/skill-registry.md and an external memory store via the mem_save tool. This behavior can lead to the exposure of private or sensitive global configurations to a shared project environment or external services.
  • [INDIRECT_PROMPT_INJECTION]: The skill serves as a data aggregator for sub-agent instructions, reading untrusted input from external files including SKILL.md, CLAUDE.md, and .cursorrules to generate compact rules for downstream agents.
  • Ingestion points: Reads files from project root and various global directories in Step 1 and 2.
  • Boundary markers: The skill does not implement delimiters or warnings to ignore embedded instructions within the summarized content.
  • Capability inventory: File system reading, local file writing, and usage of the mem_save tool for external persistence.
  • Sanitization: There is no evidence of escaping, validation, or filtering of the content extracted from the external source files before it is processed into the registry.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 12:24 AM
Security Audit — agent-trust-hub — skill-registry