goal-prompt-crafter
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest user requirements (untrusted data) to generate structured "goal" prompts and "execution briefs" for autonomous agents. There is no explicit sanitization or boundary marking (e.g., "ignore embedded instructions") mentioned for the user-provided task description, which could allow a user to inject malicious instructions into the final goal file.
- Ingestion points: User input defining the task, verification methods, and constraints (Step 1 & 2).
- Boundary markers: Absent in the generated output structure.
- Capability inventory: The skill utilizes file-writing capabilities to save generated goals and instructs the agent to provide commands (like /goal) that trigger autonomous loops.
- Sanitization: Not explicitly implemented in the workflow.
- [COMMAND_EXECUTION]: The skill instructions require the agent to write content to the file system (Step 5) as a standard behavior. It specifies creating files named
goal-<slug>.mdin the user's working directory. While this is a standard developer workflow, it involves file system modification based on content derived from user input.
Audit Metadata