goal-driven
Pass
Audited by Gen Agent Trust Hub on Apr 17, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill uses a template to construct instructions for a subagent, which includes placeholders for user-defined goals and criteria. While this is the intended functionality, it creates a surface for indirect prompt injection if the input provided by the user contains malicious instructions meant to hijack the subagent's behavior.
- Ingestion points: User input for goal and criteria definitions in SKILL.md.
- Boundary markers: The subagent prompt uses XML-style tags (, ) to delimit user input.
- Capability inventory: Uses the spawn_agent tool to delegate tasks.
- Sanitization: No explicit sanitization of user-provided goals or criteria is performed before interpolation into the subagent prompt.
- [COMMAND_EXECUTION]: The skill encourages the agent to use verification methods such as tests, logs, and command outputs to evaluate success criteria. This is part of the standard operational workflow for development-oriented tasks and does not involve arbitrary or hardcoded dangerous commands.
- [REMOTE_CODE_EXECUTION]: The README references external GitHub repositories as examples of projects built using this method. These are provided as legitimate examples of complex engineering tasks (a TypeScript compiler and a SQLite implementation) and do not represent a remote code execution risk within the skill's own logic.
Audit Metadata