intake-interview-discipline
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command (
npx skills add FuDesign2008/open-skills -g --skill <name> --yes) if required dependencies are missing. This involves shell invocation of a Node.js-based management tool.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates the download of external code and configurations from the author's repository (FuDesign2008/open-skills) during the dependency resolution phase at startup.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input during the intake process, which is then interpolated into host contracts and completion reports. This creates an attack surface where instructions hidden in task goals could influence agent behavior.\n - Ingestion points: User-provided 'Destination' descriptions and clarified responses to interview questions within SKILL.md.\n
- Boundary markers: The skill uses markdown templates from reference.md to structure data but does not specify explicit delimiters or 'ignore instructions' warnings for user content.\n
- Capability inventory: The skill is designed to guide 'goal-driven hosts' which typically possess extensive tool-execution and file-system capabilities.\n
- Sanitization: No explicit input validation or instruction-filtering logic is documented in the skill scripts or instructions.
Audit Metadata