find-skills
Warn
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Retrieves search results and skill metadata from 'https://skills.sh/api/search'. It also uses the 'web_fetch' tool to scrape security audit data from 'https://skills.sh/audits'.
- [REMOTE_CODE_EXECUTION]: Fetches 'SKILL.md' instruction files from arbitrary GitHub repositories using 'raw.githubusercontent.com' and installs them into the agent's environment via the 'learn_skill' tool. This allows for the dynamic acquisition of new capabilities from external sources.
- [COMMAND_EXECUTION]: Executes 'curl' commands to interact with external APIs and download content from the web.
- [PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection by ingesting and 'learning' instructions from external sources.
- Ingestion points: Fetches external content from 'skills.sh' and 'raw.githubusercontent.com' (SKILL.md).
- Boundary markers: Lacks technical delimiters for the 'learn_skill' payload, though it includes a manual behavioral review requirement for the agent.
- Capability inventory: Includes 'curl', 'web_fetch', 'learn_skill', and 'uninstall_skill'.
- Sanitization: Relies on manual verification and basic checks for empty content rather than technical escaping or automated safety filtering.
Audit Metadata