loop-goal
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied actions and goals to generate a /loop command. This architecture presents an indirect prompt injection surface if the input data (such as a goal description) is sourced from untrusted external data like pull request comments or public bug reports.
- Ingestion points: User-provided 'Ação', 'Goal', and 'Métrica' fields in SKILL.md templates.
- Boundary markers: The templates interpolate these values directly into the final command without explicit delimitation or 'ignore instructions' markers.
- Capability inventory: The generated /loop command allows the agent to repeatedly call tools (file system, command line, network) until a threshold is met.
- Sanitization: The skill enforces a measurable structure but does not sanitize the content for malicious instructions.
- [DYNAMIC_EXECUTION]: The skill functions as a code generator, producing complex /loop commands based on predefined templates and user input.
- Patterns: Template-based prompt generation used to construct the logic for the agent's autonomous execution loop.
Audit Metadata