The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill is designed to ingest and process external data, specifically previous meeting minutes (minutes_from) and agenda items. These inputs represent an attack surface where a malicious user or document could embed instructions aimed at overriding the meeting's procedural rules or the agent's reasoning during the debate stage.
Ingestion points: YAML fields minutes_from and agenda in MEETING.yml or similar meeting files.
Boundary markers: None identified. The skill lacks explicit instructions to ignore embedded commands within the ingested data.
Capability inventory: The skill is permitted to use read_file and write_file. While it lacks network or code execution capabilities, an injection could still result in unauthorized file modifications (e.g., falsifying meeting minutes) or influencing downstream decision-making.
Sanitization: No evidence of input validation or sanitization for the content of meeting minutes or agenda items.

roberts-rules