Agent State Integrity
Status: live. Write-time and read-time state controls ship in the platform; the product page is griff.run/product/agent-state-integrity.
Page updated 2026-08-10 against the live platform.
Reading something is not permission to remember it. Agent State Integrity gates durable agent state — what an agent is allowed to turn into lasting memory or authority — on five questions, asked in order:
| Gate | Question |
|---|---|
| 1. Evidence exists | Does the cited reference resolve to a real record? |
| 2. Evidence supports the claim | A real source is not enough — does it actually ground what the agent wants to persist? |
| 3. Source authority | May this source create this kind of state? A webpage may inform a summary without gaining authority to install policy. |
| 4. Writer authority | May this writer make the transition? Tool access is not permission; the authenticated agent or person needs the right to persist it. |
| 5. Safe at read | Every later read rechecks lineage, revocation, expiry, scope, and the purpose of the current task. |
What this gives you
- Write-time controls — candidate memories that fail a gate are observed, never promoted to future authority.
- Read-time admissibility — admitted state is rechecked when used, so revoked or expired state stops mattering immediately.
- Quarantine, revocation, and downstream-use tracing — when state is pulled, you can see where it already flowed.
- Evidence-chain receipts — every admitted write and later use carries a receipt.
Plan availability
State Integrity policy simulation is included from the Starter plan; write-time and read-time enforcement controls are part of the Team plan and above (see griff.run/pricing).