Compliance path

Who this path is for

You need guardrails, approvals, and secrets handled before agents touch production data.

This is a lighter compliance route — not legal advice — focused on Praison checks and govern steps.

Suggested order

1. Security — Untrusted input: treat user text as untrusted.

2. Security — Block bad output: stop unsafe replies at the boundary.

3. Reliability — Check the output and Guardrail in code: repeatable checks in code.

4. Security — Approve a tool run: human or policy gate before shell or file tools.

5. Security — Secrets in env: keys outside source control.

6. Security — Run tools sandboxed: limit what tools can reach.

7. Agent lifecycle — Govern: map approvals and secrets to the govern phase.

8. Security — Security quiz: confirm the sequence.