The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/bot_send.py attempts to load environment variables from hardcoded absolute paths: /Users/server/ai_projects/telegram_agent/.env.local and /Users/server/ai_projects/telegram_agent/.env. These files typically store sensitive API keys and secrets.\n- [DATA_EXFILTRATION]: The skill script scripts/bot_send.py accesses a local SQLite database at a fixed absolute path: /Users/server/ai_projects/telegram_agent/data/telegram_agent.db. Accessing localized sensitive data combined with the skill's ability to send network requests via the Telegram API creates a potential path for data exfiltration.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from Telegram messages and local markdown drafts.\n
Ingestion points: Messages are retrieved via scripts/telegram_fetch.py and local drafts are parsed in the publish_draft function.\n
Boundary markers: There are no explicit instructions or markers used to isolate external content from the agent's control instructions.\n
Capability inventory: The skill can perform network operations (sending messages, downloading media) and modify the local filesystem (updating Obsidian notes, moving files).\n
Sanitization: The skill uses basic HTML escaping in bot_send.py and regex-based markdown conversion in telegram_fetch.py, but lacks robust protection against prompt injection attacks hidden in message content.\n- [EXTERNAL_DOWNLOADS]: SKILL.md references an external repository (https://github.com/glebis/telegram_dl) for authentication and setup. This resource belongs to the skill's author.

telegram