protocol-browser-anti-stall

Pass

Audited by Gen Agent Trust Hub on Oct 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches and executes the @playwright/cli package from the official npm registry via npx. This is a standard procedure for using Playwright's automation tools.
  • [DYNAMIC_EXECUTION]: The skill utilizes run-code and eval commands to run JavaScript within the browser environment. These are used for inspecting page state and waiting for elements, and the instructions explicitly limit their use to read-only actions to prevent unintended side effects.
  • [INDIRECT_PROMPT_INJECTION]: The skill interacts with external websites, creating an inherent surface for indirect prompt injection. 1. Ingestion points: Data is gathered from browsers using snapshot, screenshot, console, and requests commands in SKILL.md. 2. Boundary markers: The skill does not define specific output delimiters for tool results, relying instead on agent reasoning and observation protocols. 3. Capability inventory: Interaction capabilities include click, type, fill, run-code, and eval as defined in references/mcp-to-cli-map.md. 4. Sanitization: Safety is managed through instructional constraints, such as limiting execution to inspection-only tasks and requiring visual evidence before retrying actions.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 19, 2026, 06:58 PM