potion-yjs-dev-browser-test

Warn

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to dynamically generate and execute shell scripts using the dev-browser utility to control a persistent Chrome instance.
  • Evidence: "For each new case, write a small one-off dev-browser --connect http://127.0.0.1:9222 <<'EOF' script from the current repro steps."
  • [PROMPT_INJECTION]: The skill reads and processes content from external web pages (Potion documents), creating a surface for indirect prompt injection where malicious content in a document could influence the agent's behavior during a test session.
  • Ingestion points: Browser pages accessed via browser.getPage and document state reading using selectors like [data-slate-editor="true"] (SKILL.md).
  • Boundary markers: None observed; the instructions do not specify delimiters or warnings to ignore embedded instructions in the ingested document content.
  • Capability inventory: Shell script execution via dev-browser, network condition manipulation via Chrome DevTools Protocol (CDP).
  • Sanitization: No sanitization or validation of external content is specified before the agent processes the document state.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 2, 2026, 05:10 PM
Security Audit — agent-trust-hub — potion-yjs-dev-browser-test