recipe-fit-assessor

Pass

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill is restricted to a planning-only (L0) capability tier, with explicit prohibitions against executing Python code, Blender, or any runtime operations. It maintains strict 'Not Run' and 'Not Produced' truth states for all assessments.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied goal summaries and deliverable targets, creating a surface for indirect instructions. However, this is inherent to its primary purpose and the risk is mitigated by the lack of any executable tools or network access.
  • [INGESTION_POINTS]: User-provided fields 'user_goal_summary' and 'target_output_destination' in contract.json.
  • [BOUNDARY_MARKERS]: A structured decision tree and specific output templates in SKILL.md guide the agent to remain within a specific assessment framework.
  • [CAPABILITY_INVENTORY]: The skill has no dangerous capabilities; it is explicitly limited to planning and prohibited from performing file writes, network operations, or subprocess execution.
  • [SANITIZATION]: No explicit input sanitization or escaping mechanisms are defined, as the skill relies on structural constraints and agent instruction adherence within a text-only mode.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 16, 2026, 03:14 PM