skills/astro-han/skills/goal-writer/Gen Agent Trust Hub

goal-writer

Pass

Audited by Gen Agent Trust Hub on Jul 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill directs the agent to execute a local Python script (scripts/lint_goal.py) to validate generated goal contracts. The script uses standard libraries to enforce character limits and identify placeholders.\n- [PROMPT_INJECTION]: The instructions command the agent to "Preserve the requested outcome" and explicitly state to "not invent limits or substitute a safer intermediate objective." This is a directive to override internal safety heuristics that might otherwise suggest safer alternatives to a user's stated goal.\n- [INDIRECT_PROMPT_INJECTION]: The skill facilitates a workflow where untrusted user intent is transformed into a contract and subsequently processed by a script. It lacks security boundaries or sanitization steps.\n
  • Ingestion points: User-provided "rough intent" processed according to instructions in SKILL.md.\n
  • Boundary markers: None identified in the prompt instructions to delimit user-provided content.\n
  • Capability inventory: Execution of the lint_goal.py script which includes file system read access.\n
  • Sanitization: No sanitization or escaping of user-provided content is performed before processing or linting.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 13, 2026, 03:00 AM
Security Audit — agent-trust-hub — goal-writer