craft-context-engineering
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill features a "CRAFT Audit Mode" designed to analyze and diagnose user-supplied prompts for structural completeness. While this constitutes an ingestion surface for untrusted data, the skill's instructions are focused on structural analysis rather than obedience to the content of the ingested prompt.
- Ingestion points: User prompts pasted into the agent for auditing (SKILL.md).
- Boundary markers: Absent for the specific ingested prompt, though the framework itself imposes a structural boundary (CRAFT elements).
- Capability inventory: The skill utilizes standard tools for factual verification (web_search, web_fetch) as defined in the referenced
operating-principles.md. - Sanitization: Not explicitly performed on the input prompt; however, the skill is constrained by strict "No-Harmful-Output" and "Fact-Check" rules in the core principles.
- [SAFE]: The skill incorporates mandatory reference files (
operating-principles.mdandplugin-specific-rules.md) that establish high security and accuracy standards. These include a 4-tier source hierarchy for verification, a strict question budget to prevent excessive interaction, and a hard prohibition on fabricating business metrics or generating non-English content that cannot be verified.
Audit Metadata