Guides · APRF practice notes
Agents Guides
Bound autonomy and require humans for high-impact actions. These guides support APRF Agent Governance, Human Approval, and Tool Safety.
Autonomy bounds, A2A trust, and human oversight — APRF Agents & Autonomy.
For eval gates and versioned prompts, see AI Lifecycle guides.
- Agent Autonomy Limits and Kill Switches
Unbounded agents loop, spend, and escalate. Production agents need charters, autonomy caps, and a real kill switch—not hope.
- Gate Secret File Reads for Coding Agents
If your IDE agent can read ~/.ssh, .env, cloud credential files, or kubeconfig without an extra gate, MCP metadata and docs injection become credential theft. Platform policy must decide—not the model.
- Human Approval for High-Impact AI Tool Calls
If an agent can refund customers, delete data, or page production without a human in the loop, you are past APRF Human Approval expectations for high-impact actions.
- MCP and Tool Allowlists in Production AI
MCP and function calling are powerful—and dangerous when every tool is enabled. Production agents need an allowlist, argument validation, and server-side authorization.
- MCP Tool Metadata Integrity After Approval
Install-time review of an MCP server is not a lasting trust decision. Tool descriptions and prompts can flip after a few benign calls—fingerprint approved metadata and block or re-approve on drift.
Next: Agent Governance
Open the related pillar specification for mandatory checks, artifacts, and pass conditions. Self-attest is optional.