#runtime-pre-action#runtime-evaluate#gates#architecture

Runtime evaluation & policy

Pre-action evaluation explained

The pattern that lets organizations approve, block, or allow agent side effects before they reach production systems—powered by Auctra evaluateAction.

May 10, 2026 · 8 min read · Markdown version

The decision point

Pre-action evaluation intercepts structured intent after planning but before execution. The authority layer returns allowed, blocked, or require_approval with an action_request_id for audit correlation.

Auctra evaluateAction implements this pattern with REST and TypeScript SDK clients. Call it from agent frameworks, MCP servers, or custom orchestration layers.

Payload design

The SDK requires agentId and actionType; place fields such as target, amount, currency, and environment inside payload. Pass a stable idempotencyKey as the second evaluateAction option when retrying the same logical action.

Avoid sending raw model transcripts as policy inputs; structure beats prose for compliance. Consistent schemas across agents simplify policy authoring.

Latency and availability

Authority checks should be fast enough for interactive agents and reliable enough for batch workflows. Cache delegation reads locally only when you can tolerate brief staleness; financial paths should always evaluate fresh.

When Auctra is unreachable, choose fail-closed for monetary actions and document the fallback in your runbook. Never silently bypass evaluation for convenience.

Rollout strategy

Shadow mode: log evaluateAction decisions without enforcing blocks during week one. Enforce on Business-critical paths first, then expand action coverage.

Auctra ties sponsors, expiring delegations, and pre-action evaluation into one accountability chain your security and finance teams can audit.

Key takeaways

  • Authority is enforced before side effects — use Agents registry and evaluateAction together.
  • Every production agent needs a named sponsor and bounded delegation visible in the console.
  • Blocked and approval-required outcomes are evidence, not failures — review them in Agents registry.

Implementation checklist

  1. Sign up at console.auctra.tech and open Agents registry (/console/agents).
  2. Register one agent with a named human sponsor accountable for its actions.
  3. Create a narrow delegation aligned with this article's workflow (Pre-action evaluation explained).
  4. Call evaluateAction from your agent or SDK before the consequential tool executes.
  5. Confirm sponsor, delegator, decision, and outcome appear in Audit or Agents registry.

People also ask

What is pre-action evaluation?
A mandatory authority check on structured agent intent immediately before the downstream system executes a consequential operation.
When should pre-action evaluation run?
Before any action that moves money, changes customer records, sends external communications, or triggers irreversible operations.
How does Auctra help with agent authority?
Auctra registers sponsors, issues expiring delegations, evaluates actions before execution, and preserves auditable accountability records.

Try in Auctra Console

Maps to: Agents registry

Pilot runtime pre action in Auctra Console

Use Agents registry to apply this guide — register an agent, delegate authority, evaluate one real action, and inspect the audit trail. Free on Builder.

  1. Create a free account: https://console.auctra.tech/auth/signup?utm_source=blog&utm_medium=cta&utm_campaign=pre-action-evaluation-explained
  2. In Agents registry (https://console.auctra.tech/console/agents), run a free Builder pilot for one production workflow.
  3. Issue a bounded delegation with limits and expiration matching this guide.
  4. Integrate evaluateAction (SDK or REST) before money, data, or infrastructure changes execute.
  5. Open Audit to verify sponsor, delegator, reviewer, and decision are recorded.

Part of guide

Runtime evaluation & policy

Pre-action gates, policy engines, approval routing, and the difference between observing agents and governing side effects.

Browse full guide →

Related guides

Make authority executable.

Evaluate agent actions against bounded, expiring delegation before they reach the real world. Start free on Builder — upgrade when audit retention and accountability matter.