The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through data ingestion.
Ingestion points: External content is ingested from the X platform via fxtwitter API (in scripts/archive_x.py) and from Telegram Saved Messages via the Telethon library (in scripts/sync_telegram_saved.py).
Boundary markers: The skill uses Markdown headers (e.g., ## Summary) and HTML comments (e.g., ) as placeholders for the AI to fill. While these provide structure, they do not strictly prevent the AI from following instructions embedded in the ingested content.
Capability inventory: The skill has the capability to write files to the user's Obsidian vault and execute local Python scripts via subprocess.run (in scripts/sync_telegram_export.py and scripts/sync_telegram_saved.py).
Sanitization: The scripts perform sanitization for filenames and metadata fields, but the actual content of the tweets or articles is processed as raw text for summarization and translation, providing a surface for adversarial content to influence the agent's output.
[COMMAND_EXECUTION]: The synchronization scripts scripts/sync_telegram_export.py and scripts/sync_telegram_saved.py utilize subprocess.run to call the companion script archive_x.py. While the inputs are passed as an argument list which mitigates shell injection risks, this pattern allows the execution of local files based on parameters derived from external data (URLs).
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch data from api.fxtwitter.com and Telegram's MTProto API. These operations are core to the skill's intended purpose of archiving remote social media content.

soia-pkm-clip-x