The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to capture live authentication credentials from the user's active session, specifically identifying the 'authorization' Bearer token and the 'x-csrf-token' (derived from the 'ct0' cookie).
[COMMAND_EXECUTION]: The skill relies on the playwriter CLI to execute arbitrary Javascript blocks. These scripts have access to the Node.js runtime (e.g., using node:fs to read local files) and the browser context (e.g., using fetch to send network requests).
[DATA_EXFILTRATION]: The instructions describe setting up a request listener in the browser to log url, method, and postData for network traffic. This monitoring could capture sensitive data from various requests made within the user's active session during the 'exploration' phase.
[PROMPT_INJECTION]: The skill features a surface for indirect prompt injection by processing external markdown articles for conversion and upload.
Ingestion points: Markdown content is read from local files such as ./tmp/x-article-content-state.json.
Boundary markers: None; the instructions do not specify delimiters or warnings for the agent to ignore embedded instructions in the article content.
Capability inventory: The skill can read local files, execute shell commands via playwriter, and perform authenticated network requests to x.com.
Sanitization: No sanitization or validation of the article content is described before it is parsed and used to build the API payload.

x-articles