The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by fetching untrusted content from Telegram chats and displaying it to the agent.\n
Ingestion points: The recent, search, unread, and thread commands in scripts/telegram_fetch.py fetch message data from Telegram chats.\n
Boundary markers: Messages are formatted using markdown blockquotes in the format_output function, providing some structural separation but not preventing adversarial injection.\n
Capability inventory: The agent can send messages, send files, download media, and publish drafts (which involves moving and modifying files in the vault).\n
Sanitization: Raw message content is presented without escaping or filtering.\n- [DATA_EXFILTRATION]: The skill accesses sensitive configuration files and user data required for its primary functions.\n
Sensitive paths: Accesses ~/.telegram_dl/config.json and ~/.telegram_dl/user.session which contain Telegram API credentials and session tokens.\n
Vault access: Accesses and modifies files within the ~/Brains/brain directory (Obsidian vault) during the draft publishing process.\n- [COMMAND_EXECUTION]: The skill executes local Python scripts to interact with the Telegram API and manage local files.\n
Scripts: scripts/telegram_fetch.py and scripts/bot_send.py perform network requests to Telegram servers and modify the local file system (downloads, vault updates).

telegram