The Agent Skills Directory

[COMMAND_EXECUTION]: The file upload handler generates a task description that instructs the agent to use the Bash tool to execute curl and python commands. These commands are constructed by interpolating external file paths and environment variables into the shell string, which presents a risk of command injection if the input is not perfectly sanitized.
[DATA_EXFILTRATION]: The /logs and /memory commands allow authorized Telegram users to retrieve content from internal system files, specifically session-gap-log.jsonl and learnings.md. This mechanism can be used to export sensitive project history or previous agent learnings to an external platform.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting untrusted text and file content from Telegram and passing it to sub-agents. While it uses <untrusted_telegram_*> delimiters and explicit instructions to mitigate risks, the agent's behavior could still be influenced by malicious content embedded in user messages or uploaded files.
[EXTERNAL_DOWNLOADS]: The skill downloads user-provided files from the Telegram Bot API (api.telegram.org). While this is a well-known service, the skill initiates these downloads based on external triggers.

telegram-polling