The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes untrusted message content from external Telegram sources.
Ingestion points: scripts/client.py fetches message data from Telegram chats and channels using the telethon library.
Boundary markers: No delimiters or specific instruction-isolation markers are implemented to separate untrusted message content from the system context.
Capability inventory: The skill can read from and write to the Telegram account and access a local SQLite database, but it does not include direct shell execution or arbitrary file system write tools.
Sanitization: There is no evidence of message content sanitization or filtering to prevent embedded instructions from influencing the agent's behavior.
[CREDENTIALS_UNSAFE]: Hardcoded public API credentials are present in scripts/config.py. The TG_API_ID (2040) and its associated api_hash are included as a fallback. These are the publicly known credentials for the official Telegram Desktop application. The code includes a warning and instructions for users to provide their own private credentials via environment variables to minimize account risks.

tg-hub