The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the tg-kit utility from the author's GitHub repository using the uv tool manager. This is a core part of the skill's functionality to enable Telegram interaction.
Evidence: command -v tg || uv tool install 'git+https://github.com/ruslan-korneev/tg-kit' in SKILL.md.
[COMMAND_EXECUTION]: The skill relies on executing the tg command-line tool for all Telegram-related operations, including reading chat history, sending messages, and calling MTProto methods.
Evidence: Extensive documentation of tg subcommands like send, read, press, and raw throughout SKILL.md and references/raw.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources (Telegram messages and bot responses), which creates a potential surface for indirect prompt injection attacks.
Ingestion points: tg read, tg msg, tg search, and tg dialogs commands in SKILL.md.
Boundary markers: The skill instructions do not specify explicit delimiters for the agent to use when interpreting message content.
Capability inventory: The skill has the ability to write files via tg download and communicate externally via tg send and other write commands.
Sanitization: The tool includes a safety guard and allowlist mechanism (tg allow) that requires explicit user confirmation before sending data to third-party peers, mitigating the risk of automated exfiltration via injection.

telegram