The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/fetch_x_content.py uses subprocess.run to call the twitter-cli tool. The implementation is secure as it passes arguments as a list and does not use a shell environment, which prevents command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill can optionally fetch data from the Jina Reader service (r.jina.ai) when structured extraction fails. This behavior is properly gated by an --allow-jina flag and the instructions explicitly require the agent to obtain user approval before sending URLs to this third-party service.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from X/Twitter posts and articles.
Ingestion points: Content is retrieved via twitter-cli or Jina Reader in scripts/fetch_x_content.py.
Boundary markers: The output uses Markdown headers (e.g., # Post text) and YAML-style frontmatter delimiters (---) to separate metadata from the external content.
Capability inventory: The skill can execute the twitter command, write output to a user-specified file path, and perform HTTP GET requests to Jina AI.
Sanitization: The script performs basic URL normalization and extracts specific data fields into a structured format to minimize the risk of the agent misinterpreting the retrieved content as instructions.
[CREDENTIALS_UNSAFE]: The skill documentation notes that twitter-cli may reuse existing local authentication tokens or browser sessions. It provides clear guidance for the agent to inform the user before the first credentialed run and strictly forbids the printing or persistence of these sensitive tokens.

x-twitter-reader