The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external data (Telegram messages) through the tg read and tg search commands. This constitutes an ingestion surface for untrusted content.
Ingestion points: Message retrieval functions in scripts/tg.
Boundary markers: The skill provides a logical protocol for Agent-to-Agent communication (A2A) with specific headers and a dedicated forum topic, though no technical isolation of message content is implemented at the script level.
Capability inventory: The CLI permits sending (tg send) and deleting (tg delete) messages, as well as managing read cursors.
Sanitization: The script performs no content sanitization; security relies on the agent following instructions in references/a2a.md to treat incoming messages as data rather than instructions.
[CREDENTIALS_UNSAFE]: The skill manages MTProto session files which grant full account access. Documentation in references/setup.md correctly identifies the sensitivity of these files and instructs users to apply restrictive permissions (chmod 600) and avoid copying them.

telegram-cli