The Agent Skills Directory

[COMMAND_EXECUTION]: The ClaudeBridge in src/telegram_telethon/daemon/claude_bridge.py spawns the claude CLI using asyncio.create_subprocess_exec. The prompt passed to the CLI via the -p flag is derived from incoming Telegram messages matched in src/telegram_telethon/daemon/runner.py. If the agent has tool access (e.g., Bash, Edit), this creates a pathway for arbitrary command execution.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection.
Ingestion points: Incoming Telegram messages monitored by the daemon in runner.py.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands within the message text passed to the LLM.
Capability inventory: The skill can execute shell commands via the claude CLI bridge and perform file operations.
Sanitization: There is no sanitization of the message content before it is interpolated into the CLI command.
[EXTERNAL_DOWNLOADS]: The transcribe_voice function in src/telegram_telethon/modules/media.py sends local audio files to api.groq.com for transcription. While Groq is a well-known service, users should be aware that their voice data is transmitted to an external provider.
[CREDENTIALS_UNSAFE]: The skill manages sensitive Telegram API credentials (api_id, api_hash). During the setup process in scripts/tg.py, these are stored in ~/.config/telegram-telethon/config.yaml. The skill correctly mitigates risk by setting restricted file permissions (0600) so only the file owner can read the secrets.

telegram-telethon