Greyline: Agent Security

Your agent calls APIs, MCP servers, web search, and email every day. Every one of those responses is an attack surface. A malicious endpoint can embed hidden instructions that hijack your agent's next action. A compromised MCP server can return data designed to rewrite your agent's priorities. A search result can contain a prompt injection that leaks your credentials.

Greyline Agent Security is the defensive layer between your agent and the hostile internet. Built from real production experience at The Meridian Lab, where we build counter-intelligence tools for the autonomous internet. The skill covers five domains: tool result validation with injection scanning and schema checking, persistent endpoint trust scoring with automatic quarantine, a structured MCP server vetting protocol, dedicated email and web content threat assessment, and five defensive operation rules (Source, Escalation, Least Action, Verification, Credential) that prevent the most common agent manipulation attacks.

Setup takes 5 minutes. Copy the SKILL.md, create a trust ledger file, add one line to your SOUL.md. Your agent immediately starts scanning tool results, building trust scores, and flagging suspicious content instead of blindly acting on it. Normal operations flow through untouched — you only notice when something tries to manipulate your agent.

Ships with: SKILL.md, TRUST_LEDGER.md template, START-HERE.md install guide, and structured alert templates for injection detection, trust score changes, and anomaly reporting.