The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file scripts/config.py contains a hardcoded _DEFAULT_API_HASH value ("b18441a1ff607e10a989891a5462e627"). While documented as a public fallback for Telegram Desktop, hardcoding cryptographic secrets is an unsafe practice.
[COMMAND_EXECUTION]: The SKILL.md file contains a specialized link (minis://open_terminal?init_command=...) designed to execute a series of shell commands in the user's terminal, including uv run python -c. This mechanism allows for arbitrary command execution if the link parameters are manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest, synchronize, and search messages from external Telegram groups and channels. This data is untrusted and could contain malicious instructions designed to influence the agent's behavior when it processes or summarizes the retrieved messages.
Ingestion points: Messages are fetched from the Telegram MTProto API and stored in a local SQLite database (scripts/client.py and scripts/db.py).
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present when processing fetched message content.
Capability inventory: The skill can read chat lists, fetch historical messages, and perform regex-based searches.
Sanitization: While standard keyword searches are handled, the search_regex method in scripts/db.py compiles raw regex patterns which could be used to trigger resource exhaustion (ReDoS) or bypass filters if the pattern is sourced from untrusted input.
[EXTERNAL_DOWNLOADS]: The skill connects to Telegram's MTProto servers to fetch data. While this is the intended purpose, it involves network operations to a third-party service.

tg-hub