Robot Validation

After each task runs, Validation Agent checks the result against the expected output and validation rules defined by the Tasks stage — and returns a structured verdict with a quality score.

This is a pipeline stage agent. It does not appear in the AI Expert chat list. You'll find it in the **Execution Pipeline** dropdown when replacing the Validation stage agent in a Robot's Advanced settings.

## What it does

- Receives a completed task definition alongside its execution result
- Evaluates whether the result satisfies the expected output
- Checks compliance with any validation rules defined for the task
- Returns a JSON verdict: passed/failed, a quality score (0.0–1.0), a list of issues, and a plain-language explanation

Validation is semantic — it reads and interprets the result, not just checking whether it exists or has a certain format.

## Scoring scale

| Score | Meaning |
|-------|---------|
| 1.0 | Perfect — meets or exceeds all expectations |
| 0.8–0.99 | Good — minor improvements possible |
| 0.6–0.79 | Acceptable — some issues present |
| 0.4–0.59 | Marginal — significant issues |
| 0.0–0.39 | Unacceptable — fails requirements |

A `passed: false` verdict means the task result fails at least one critical requirement. The pipeline uses this to decide whether to retry, escalate, or proceed.

## Capabilities at a glance

| Capability | Details |
|------------|---------|
| Agent type | `robot-validation` — pipeline stage agent |
| Replaceable | Yes — available in the Validation stage dropdown |
| Input | Task definition (instructions, expected output, validation rules) + execution result |
| Output | JSON verdict: `passed`, `score`, `issues`, `details` |
| Sandbox | No container |

## What's next

- [Replacing Stage Agents](/docs/en-us/mission-control/execution-pipeline/replacing-stage-agents) — how to swap this agent for a custom one
- [Pipeline Overview](/docs/en-us/mission-control/execution-pipeline/pipeline-overview) — how all stages connect
- [Robot Delivery Agent](/docs/en-us/built-in-agents/robot-delivery) — the final stage that formats and delivers results