Agent Security
Treats anything the agent did not receive from the user as data rather than instructions, and covers dependency and MCP supply chain vetting.
Free rules for coding agents.
Rules load in every session, so the agent applies them without being asked. These three change how it handles evidence and untrusted input: what it treats as instructions, when a measurement counts as confirmed, and what it may state as fact.
Measurement Verification
Treats a count or status read out of a tool as a claim the tool is making, requiring a second method that fails differently before it becomes a conclusion.
No Unverified Claims
Separates what the agent read or ran this session from what it recalled, and labels anything it cannot source as unverified.
Want AI that works?
84EM builds the agent skills, MCP integrations, and automations behind these. Describe the workflow you want handled.
Folks We've Helped
