The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from X/Twitter, which is a known vector for indirect prompt injection attacks where malicious instructions could be embedded in tweets to influence the agent's behavior.
Ingestion points: The scripts/scan_x.py script fetches real-time post content from X/Twitter via the xAI Grok API search tool.
Boundary markers: The prompt sent to the Grok API (in scan_x.py) lacks explicit boundary markers or instructions to treat the ingested tweet content as potentially untrusted data.
Capability inventory: The skill has permissions to read the .env file, perform network operations to the xAI API, and write digest files to the local filesystem via the scripts/scheduled_scan.sh script.
Sanitization: The script does not perform sanitization or filtering of the tweet content beyond asking the LLM to skip 'generic motivational posts' and 'pure promo'.

x-scanner