create-skill
Pass
Audited by Gen Agent Trust Hub on Jun 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes user-supplied Name, Trigger, and Behavior inputs to create new instruction sets. This creates an indirect prompt injection surface because malicious behavior descriptions could be crafted to influence or override the agent's logic when the newly created skill is later loaded.
- Ingestion points: User interaction inputs for 'Trigger' and 'Behavior' provided during the scaffolding process.
- Boundary markers: Absent; the skill does not wrap the interpolated user content in delimiters or include instructions to ignore embedded commands.
- Capability inventory: The skill performs file creation (skills//SKILL.md) and modifies the plugin configuration file (.claude-plugin/plugin.json).
- Sanitization: Absent; the user's behavior description is used directly to derive the instructions for the new skill without validation or escaping.
Audit Metadata