The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes content from external Telegram channels and discussion groups, which serves as an ingestion point for untrusted data that may contain malicious instructions targeting the AI agent.
Ingestion points: reader.py and reader_telethon.py fetch message text and user comments from Telegram.
Boundary markers: The skill lacks delimiters or specific instructions to the agent to treat fetched content as untrusted data.
Capability inventory: The skill uses network tools to connect to Telegram and can write data to local files using the --output flag.
Sanitization: Fetched message content is provided to the agent without filtering or escaping.
[CREDENTIALS_UNSAFE]: The skill operates by accessing and managing Telegram session files (.session) which grant full access to a user's account. The diagnostic utility tg-reader-check exposes the absolute paths, file sizes, and modification timestamps of these sensitive files to the agent's context.
[PERSISTENCE]: The setup-tg-reader.sh script and installation instructions recommend modifying the user's shell profile (.bashrc) to update environment paths for virtual environments.
[DYNAMIC_EXECUTION]: The tg-reader-check utility uses the __import__ function in tg_check.py to dynamically load and verify the status of dependencies like pyrogram and telethon during a health check.

sergei-mikhailov-tg-channel-reader