web-to-prd

Warn

Audited by Gen Agent Trust Hub on Apr 9, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes shell commands via the Bash tool to manage local file system state, specifically targeting lock files within the user's home directory to manage browser profiles (e.g., rm -f "$HOME/.playwright-profile/SingletonLock").
  • [REMOTE_CODE_EXECUTION]: The skill involves the dynamic execution of remote code by instructing the agent to install and run MCP servers using npx. This pattern fetches and executes packages such as @playwright/mcp and firecrawl-mcp from the npm registry at runtime.
  • [DATA_EXFILTRATION]: The skill's core functionality involves crawling potentially authenticated web application sessions and exporting the extracted metadata and screenshots to an external service (Notion). This represents a managed data exfiltration path where sensitive application data is moved to a third-party cloud environment.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from external websites and processes it to generate structured documentation.
  • Ingestion points: Raw web content, UI element text, and page snapshots retrieved via Playwright tools.
  • Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish between crawled data and the skill's operational instructions.
  • Capability inventory: Access to Bash, local file writing, web searching, and Notion API integration.
  • Sanitization: The skill does not describe any sanitization or filtering of the retrieved web content before it is used for feature extraction and PRD generation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 9, 2026, 01:06 PM