feature-generator

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill contains an inherent indirect prompt injection surface because it reads and interprets data from user-modifiable markdown files without isolating the data from instructions.\n
  • Ingestion points: SKILL.md (Step 2 and Step 3) specifies reading and extracting functionality from spec.md and features.md.\n
  • Boundary markers: Absent. The instructions do not prescribe any delimiters or guardrails to prevent the agent from executing instructions embedded within the files.\n
  • Capability inventory: Safe level capability inventory. The skill only writes output files to /mnt/user-data/outputs/features.md and /mnt/user-data/outputs/spec.md, and calls present_files. No network operations or command line capabilities exist within SKILL.md.\n
  • Sanitization: Absent. No escaping or filtering is applied to the content parsed from the markdown files before it is processed by the model.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 04:06 PM
Security Audit — agent-trust-hub — feature-generator