The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/build-analysis-eval.mjs contains a hardcoded API bearer token used to authenticate requests to X's internal GraphQL API.
[DATA_EXFILTRATION]: The dynamically generated script extracts the ct0 CSRF token from the user's browser cookies (document.cookie) to authorize session-based API calls to X's internal endpoints. While no external data transfer was detected, this constitutes access to sensitive session credentials.
[COMMAND_EXECUTION]: The workflow requires the agent to execute a local Node.js script and subsequently run arbitrary JavaScript within the browser context using the evaluate_script tool. The skill uses dynamic code generation to build the final browser payload from local templates and input parameters.
[PROMPT_INJECTION]: The skill processes external data (account handles), creating a surface for indirect prompt injection:
Ingestion points: Handles are read from references/default-handles.json and user-provided arguments.
Boundary markers: None present in the prompt generation logic.
Capability inventory: Local command execution via Node.js and browser script execution via DevTools MCP.
Sanitization: Input handles are validated against a strict alphanumeric regular expression (/^[A-Za-z0-9_]{1,15}$/) before being interpolated into the generated script string, which effectively mitigates common injection patterns.

analyze-x-subscriptions