The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts sync_telegram_export.py and sync_telegram_saved.py invoke the archive script archive_x.py using subprocess.run(). Arguments are passed as a list, which avoids shell interpolation and mitigates standard command injection risks. The execution is limited to internal project scripts within the skill's own directory.
[CREDENTIALS_UNSAFE]: The skill manages Telegram MTProto credentials, including an account session string. Security is addressed by instructing users to store these sensitive values in a private config.yml file located in the user's home directory (e.g., ~/.config/soia-skills/soia-pkm-clip-x/config.yml). This practice ensures secrets are not committed to version control or stored within the Obsidian vault itself. A dedicated setup script (generate_telegram_session.py) is provided to facilitate secure credential generation.
[PROMPT_INJECTION]: The skill processes untrusted external content from X and Telegram, creating a potential surface for indirect prompt injection. The agent is instructed to summarize, translate, and extract topics from the ingested data. While there is a risk that malicious instructions embedded in tweets or messages could attempt to influence the agent's behavior, the instructions emphasize data transformation (summary/topic extraction) rather than direct execution of the content.
[EXTERNAL_DOWNLOADS]: The skill fetches data from external APIs to perform its core functions. It utilizes the public api.fxtwitter.com and cdn.syndication.twimg.com endpoints to retrieve tweet information without requiring personal X API keys. Interaction with Telegram occurs via the official MTProto protocol using the telethon library.

soia-pkm-clip-x