orchestrator
Warn
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is configured with the
Bashtool and uses it to perform directory searches and system-level tasks, such as searching for existing skills using thefindcommand. - [REMOTE_CODE_EXECUTION]: The 'Adaptive Skill Creation' workflow programmatically generates new instruction sets (skills) and saves them to the
.claude/skillsdirectory. The requirement to 'Integrate Immediately' after creation constitutes a dynamic execution pattern where the agent follows instructions it generated at runtime. - [PROMPT_INJECTION]: The skill has a significant surface for indirect prompt injection because it synthesizes inputs from multiple 'specialist' sources and user requests to determine if new skills are needed.
- Ingestion points: Processes complex user requests and results returned by other skills or tools.
- Boundary markers: No specific delimiters or 'ignore' instructions are used when handling data from specialists before synthesis.
- Capability inventory: The agent possesses
Bash,Write, andEditcapabilities, which are used to implement the dynamically generated instructions. - Sanitization: There is no evidence of validation or sanitization of the context passed to the
skill-coachtool or the resulting generated skill content.
Audit Metadata