The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/capture.ts implements a dynamic code execution pattern using await import(resolvedPath). This is used to load and run 'pre-scripts' that the agent is instructed to write locally to handle browser interactions such as logging in or dismissing modals.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it visits untrusted external URLs and extracts text content from the page DOM. Maliciously crafted instructions within a website's content could attempt to influence the agent's subsequent actions in the PR workflow.
[SAFE]: The skill includes robust security controls for data privacy. It features automatic redaction of sensitive patterns (e.g., OpenAI/Anthropic keys, JWTs, passwords) in scripts/capture.ts before screenshots are taken, and a dedicated validation script scripts/validate-sensitive.ts to scan captured text for leaked credentials.

screengrabs