screenshot-feature-extractor
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted visual data from screenshots which may contain adversarial text designed to manipulate the agent's output or logic during the synthesis phase.
- Ingestion points: User-provided screenshot files are ingested and processed by multiple specialized agents in the parallel analysis phase.
- Boundary markers: The instructions for the UI, interaction, and business agents lack explicit delimiters or instructions to ignore embedded commands within the analyzed visual content.
- Capability inventory: The skill uses the
Tasktool for multi-agent coordination and has the ability to write files to the local filesystem (docs/plans/). - Sanitization: The skill does not sanitize the extracted 'product name' before using it to construct the final output file path (
docs/plans/YYYY-MM-DD-<product>-features.md), which could allow for path manipulation if an adversarial name is identified in the screenshot.
Audit Metadata