anycap-human-interaction

Pass

Audited by Gen Agent Trust Hub on Apr 24, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted feedback from humans or external web pages and uses it in subsequent prompts for image generation or video analysis.
  • Ingestion points: Results from anycap annotate poll (SKILL.md).
  • Boundary markers: No delimiters or ignore instructions are used when interpolating human feedback into prompts.
  • Capability inventory: The agent can execute anycap image generate and anycap actions video-read (SKILL.md).
  • Sanitization: The skill does not describe any validation or escaping of human-provided labels before they are used as instructions for AI models.
  • [EXTERNAL_DOWNLOADS]: The skill depends on the external anycap CLI tool and communicates with the anycap.ai domain to manage annotation sessions.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 24, 2026, 08:44 AM