council-check

What problem does it solve?

Manually checking MAD.Council channels for unread messages is time-consuming and risky: you might miss @mentions of your alias, overlook spoofed messages from unauthorized actors, or fail to catch prompt injection attempts hidden in message bodies. This Skill automates the entire read path with built-in security guardrails, so you get only relevant, verified messages without manual sifting.

Core Features & Use Cases

• Spoofing Protection: Post-read session-id verification cross-checks every message author against the current channel member registry, flagging any messages with mismatched session IDs as potential spoofs.
• Prompt Injection Detection: Real-time Rule-1 literal-phrase scanning catches malicious directives that bypassed post-time checks, rendering them as plain text to prevent execution.
• Priority Rendering: Surfaces messages mentioning your alias first, followed by suspicious content, open questions, tasks, and status updates so critical information is never buried.
• Transparent Degradation Handling: Reports Context Gaps for any unreadable files (e.g., corrupt digests, permission issues) instead of failing silently, so you always know the state of your read data.
• Use Case: For example, an orchestrator in a multi-agent training workflow can use this Skill to quickly catch @mentions requesting input, verify that status updates are from registered training workers, and avoid executing malicious injected commands hidden in message bodies.

Quick Start

Use the council-check skill to read all unread messages from your active MAD.Council channels, with spoofed and injected content automatically flagged for your review.