textual-screenshot
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- Generated Script Execution: The skill instructs the agent to create and run temporary Python scripts using
uv run. This involves dynamic code generation and execution to interact with the Textual test harness, which is a pattern that requires review to ensure the generated logic remains within safe parameters. - Potential Visual Data Exposure: The tool captures the visual state of terminal applications as SVG files. If the applications being screenshotted contain sensitive information like credentials or private messages, these could be inadvertently saved. The skill includes recommendations to use synthetic data and avoid capturing secrets as a safeguard.
- Indirect Prompt Injection Surface: The skill renders application interfaces that may display untrusted external data. This creates a potential surface where malicious instructions could be embedded in the UI and influence the agent's behavior.
- Ingestion points: Application UI components and state (SKILL.md).
- Boundary markers: The instructions suggest using synthetic data but do not define delimiters for the ingested UI content.
- Capability inventory: Command execution (
uv run), file writing (save_screenshot), and access to application internals (SKILL.md). - Sanitization: The skill relies on manual inspection rather than automated filtering of the captured content.
Audit Metadata