The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes brand mentions and social media content from external APIs and user exports, creating an inherent attack surface for indirect prompt injection.\n
Ingestion points: Data is fetched from Bluesky, Hacker News (Algolia), GDELT, and Tavily APIs, or provided via manual user exports in the skill instructions.\n
Boundary markers: The skill contains explicit instructions to treat every fetched mention as untrusted input and references a separate security protocol (SECURITY.md).\n
Capability inventory: The skill executes local Python scripts for data collection and utilizes an authorized proposal system to write events to the internal registry.\n
Sanitization: Instructions prevent external content from reclassifying its own triage status or modifying the agent's query architecture.\n- [EXTERNAL_DOWNLOADS]: The skill interacts with external services to gather monitoring data.\n
It uses connectors to access well-known services such as Tavily, GDELT, Bluesky, and Hacker News (Algolia Search).\n- [COMMAND_EXECUTION]: The skill performs its operations by calling local helper scripts.\n
Executes registry-events.py using an authorized operation: propose request to update the event stream.\n
Utilizes various scripts within scripts/connectors/ to interface with social media and forum APIs.

social-pulse-monitor