task-execution-engine
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes implementation tasks and acceptance criteria defined in external markdown design documents (e.g., docs/designs/*.md). \n
- Ingestion points: Files passed to the --file argument of scripts/task_manager.py and subsequently read by the agent. \n
- Boundary markers: The skill lacks explicit instructions for the agent to distinguish between task data and potential malicious instructions embedded within the markdown tasks. \n
- Capability inventory: The agent is authorized to write files, modify source code, and execute management scripts to fulfill the implementation loop. \n
- Sanitization: No sanitization or validation of the task descriptions or criteria is performed before the agent treats them as actionable instructions. \n- [PROMPT_INJECTION]: The SKILL.md defines 'Unattended Mode Rules' that explicitly instruct the agent to avoid stopping for questions or asking for clarification, and to make autonomous decisions. This instruction overrides standard conversational safety patterns and human-in-the-loop oversight, potentially causing the agent to execute malicious or destructive tasks found in design documents without user review.
Audit Metadata