agent-skill-discovery
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is a discovery-only utility that scans for metadata in existing configuration files. It follows a 'Read-only' principle, explicitly stating in its rules that it must never execute any discovered resource or perform recommendations.
- [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it ingests untrusted data from the local repository to generate its catalog.
- Ingestion points: The skill reads
plugin.json,SKILL.md, and.mcp.jsonfiles from the current working directory to extract metadata like names and descriptions. - Boundary markers: Discovered data is structured into a clean Markdown catalog with specific headers for global and local scopes.
- Capability inventory: The skill uses
Glob,Read, andToolSearch. It lacks capabilities for network exfiltration, arbitrary command execution, or file writing. - Sanitization: Instructions specify the use of safe parsing methods, JSON/YAML validation, and graceful skipping of malformed files.
Audit Metadata