The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Chrome DevTools Protocol (CDP) Runtime.evaluate method to execute large, dynamically generated JavaScript blocks within the context of the user's browser.
Evidence: scripts/fetch_douyin_comments.py and scripts/fetch_douyin_comments_cdp.mjs both construct and send JS strings to the browser via the bridge or direct websocket connection.
[PRIVILEGE_ESCALATION]: By connecting to the Chrome DevToolsActivePort, the skill gains full programmatic control over the browser session. This includes the ability to read all page content, access cookies, and perform actions as the logged-in user.
Evidence: scripts/douyin_cdp_bridge.mjs searches for and reads the DevToolsActivePort file to locate the browser's debugging websocket.
[DYNAMIC_EXECUTION]: The douyin_cdp_bridge.mjs script creates a local HTTP server that exposes an /eval endpoint, which accepts and executes arbitrary JavaScript code in the browser without authentication.
Evidence: The /eval route in scripts/douyin_cdp_bridge.mjs passes the expression body directly to Runtime.evaluate.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest untrusted data (Douyin comments) from the web.
Ingestion points: Douyin comment API responses fetched via the browser (scripts/fetch_douyin_comments.py).
Boundary markers: None identified. The content is saved directly to JSON/JSONL files.
Capability inventory: The agent has the ability to write to the local filesystem, spawn processes (douyin_cdp_bridge.mjs), and execute code in the browser.
Sanitization: No sanitization of comment text is performed before saving to local files, creating a risk if the agent later reads these files as instructions.

douyin-comments