skill-edit
Pass
Audited by Gen Agent Trust Hub on Apr 5, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches content from GitHub repositories (specifically raw.githubusercontent.com) to perform version comparison and updates for imported skills.
- [COMMAND_EXECUTION]: Modifies and deletes files and directories within the local environment and executes internal audit and sync tools as part of its standard lifecycle.
- [PROMPT_INJECTION]: Contains surface area for indirect prompt injection when reading existing skill files, which is managed by the following evidence chain:
- Ingestion points: Target SKILL.md and README.md files are read into context.
- Boundary markers: Explicit instructions direct the agent to treat processed skill content exclusively as data and never as instructions.
- Capability inventory: The skill can perform file write/delete operations and call internal synchronization tools.
- Sanitization: Human-in-the-loop verification is mandatory for all modification plans and irreversible deletions.
Audit Metadata