The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains a hardcoded authentication token in scripts/constants.ts: DEFAULT_BEARER_TOKEN = "Bearer AAAAAAAAAAAAAAAAAAAAANRILgAAAAAAnNwIzUejRCOuH5E6I8xnZz4puTs%3D1Zv7ttfk8LF81IUq16cHjhLTvJu4FA33AGWWjCpTnA".
[CREDENTIALS_UNSAFE]: The skill accesses sensitive environment variables X_AUTH_TOKEN, X_CT0, X_GUEST_TOKEN, and X_TWID in scripts/cookies.ts to manage X session authentication.
[DATA_EXPOSURE]: The skill reads and writes sensitive session cookies to local JSON files at ~/Library/Application Support/tuzi-skills/x-to-markdown/cookies.json or ~/.local/share/tuzi-skills/x-to-markdown/cookies.json as seen in scripts/cookie-file.ts and scripts/paths.ts.
[COMMAND_EXECUTION]: The skill launches the system's web browser process using child_process.spawn with a custom user data directory to extract cookies via the Chrome DevTools Protocol (CDP) in scripts/cookies.ts.
[COMMAND_EXECUTION]: The skill executes shell commands using execSync to interact with the Windows environment (via WSL) in scripts/paths.ts, running cmd.exe /C "echo %USERPROFILE%" and wslpath.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from X (tweets and articles) which is then converted to Markdown. This creates a vulnerability surface for indirect prompt injection as the agent is expected to read the resulting files.
Ingestion points: Fetches content from x.com and twitter.com in scripts/graphql.ts.
Boundary markers: No explicit instructions are provided to the agent to treat the converted Markdown content as untrusted data.
Capability inventory: The skill possesses capabilities for writing files (scripts/main.ts), executing shell commands (scripts/paths.ts), and launching browser processes (scripts/cookies.ts).
Sanitization: Content is escaped for Markdown formatting, but the underlying text is processed without safety filtering for potential instructions.

tuzi-danger-x-to-markdown