The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from user exports and third-party connectors, which presents a surface for indirect prompt injection.
Ingestion points: The skill ingests user-provided native analytics exports from various social platforms and data from connectors like discourse.py (forum JSON) as specified in the Data Sources section of SKILL.md.
Boundary markers: The instructions include a dedicated security warning stating that every export and connector pull must be treated as untrusted input and that their contents should be treated as data, not instructions.
Capability inventory: The skill has permissions to read and write to the memory/ file system and can propose events to a registry via the registry-events.py script as described in the Instructions.
Sanitization: The skill relies on natural language prompting to ensure the agent ignores instructions embedded in data; no explicit technical sanitization or schema validation steps are detailed in the documentation.

social-measurement-loop