goal-loop
Pass
Audited by Gen Agent Trust Hub on Jul 9, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface Analysis (Category 8):
- Ingestion points: The skill templates (e.g., in SKILL.md) instruct agents to read from potentially untrusted project-specific files like
PLAN.mdor issue tracker descriptions specified in the 'Read first' field. - Boundary markers: Absent; the suggested goal-loop templates do not explicitly include delimiters or instructions to ignore embedded commands within the ingested data.
- Capability inventory: The agent is tasked with executing shell commands (
pytest,pnpm test) and performing file system modifications (code migrations and documentation updates). - Sanitization: Absent; the skill defines a structure for human-authored prompts rather than automated data processing logic.
- [COMMAND_EXECUTION]: The skill documentation provides examples of shell-based validation commands (e.g.,
pytest -q,pnpm test,pytest --cov) for the agent to execute. These are standard development tools intended for verifying the success of autonomous tasks.
Audit Metadata