blender-recipe-decomposer
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acts as a text-processing layer that ingests "Full user intent" and interpolates it into a structured recipe table for downstream skills. This creates a standard surface for indirect prompt injection common to all text-processing skills.
- Ingestion points: The
SKILL.mdandcontract.jsonfiles define "Full user intent" (also referred to asmulti_deliverable_request) as the primary input for decomposition. - Boundary markers: The skill does not define specific delimiters or "ignore instructions" markers when formatting the user intent into the output recipe table.
- Capability inventory: The skill is categorized as a process/discipline skill with a capability tier of L0. It lacks shell execution, file system write access, or network capabilities.
- Sanitization: No explicit sanitization or filtering of user input is performed before it is passed to subsequent skills like
intent-to-3d-brief-writer. - [NO_CODE]: The skill package is composed entirely of instructional Markdown, reference documentation, and JSON configuration. No executable scripts, binaries, or third-party code dependencies are included in the distribution.
Audit Metadata