anycap-human-interaction
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to ingest and act upon feedback provided by humans through annotations and video recordings, creating a surface where malicious instructions could be embedded in visual or audio data.
- Ingestion points: AnyCap annotation sessions, recording files (.webm), and annotated screenshots (.png) retrieved via anycap annotate poll (SKILL.md).
- Boundary markers: The instructions lack specific delimiters or guardrails to treat human-provided feedback as potentially untrusted instructions.
- Capability inventory: The skill possesses capabilities to read files, execute shell commands (anycap CLI), and modify project code or generated assets based on the feedback received.
- Sanitization: There is no evidence of sanitization or filtering of the content extracted from the human interaction before it is processed by the AI.
Audit Metadata