> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.middesk.com/work-with-agents/take-action/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.middesk.com/_mcp/server.

# Take action

> Drive meaningful action based on agent findings and recommendations.

Agents are designed to work alongside you, not replace your judgment. When an agent completes its analysis, it hands off to you with structured findings and recommendations. From there, you review the results, provide input, and direct the agent to carry out the next steps. This interaction model keeps you in control of your decisions while reducing the manual effort to reach them.

## Interacting with agents

Depending on the use case, you can require an agent to pause and wait for input using a Human-in-the-loop (HITL) pattern. To support this, agents support interrupts.

> **Info**
>
> An interrupt is a [step](/reference/run) with type `interrupt`. It signals that the agent has paused execution and is waiting for input before continuing. The interrupt step includes the agent's findings as [artifacts](/work-with-agents/run-agents#viewing-agent-activity) with confidence scores and source references.

## Responding to agents

<title id="md-action-title">
  Agent Action Flow
</title>

<desc id="md-action-desc">
  A user runs an agent which plans and executes. The agent interrupts
  with sources to add, the user approves, then the agent adds sources,
  refreshes the report, and approves the business.
</desc>

<defs>
  <path d="M 0 0 L 10 5 L 0 10 z" fill="var(--md-sky)" />

  <path d="M 0 0 L 10 5 L 0 10 z" fill="var(--md-violet)" />

  <path d="M 0 0 L 10 5 L 0 10 z" fill="var(--md-success)" />

  <filter id="md-card-shadow-af" x="-6%" y="-6%" width="116%" height="120%" />
</defs>

<rect x="30" y="0" width="220" height="700" rx="12" />

<rect x="390" y="0" width="220" height="700" rx="12" />

<g transform="translate(40, 20)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="40" rx="6" />

  User
</g>

<g transform="translate(400, 20)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="40" rx="6" />

  Agent
</g>

<g transform="translate(40, 80)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="68" rx="10" />

  <text x="20" y="22">
    Client
  </text>

  <text x="20" y="48">
    Run an Agent
  </text>
</g>

<rect x="255" y="104" width="130" height="20" rx="4" />

Creates agent run

<g transform="translate(400, 80)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="68" rx="10" />

  <text x="20" y="22">
    Agent
  </text>

  <text x="20" y="48">
    Plan
  </text>
</g>

<g transform="translate(400, 200)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="68" rx="10" />

  <text x="20" y="22">
    Agent
  </text>

  <text x="20" y="48">
    Execute
  </text>
</g>

<rect x="260" y="212" width="120" height="44" rx="4" />

Interrupts with

sources to add

<g transform="translate(40, 200)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="68" rx="10" />

  <text x="20" y="22">
    Client
  </text>

  <text x="20" y="48">
    Approves sources
  </text>
</g>

<rect x="255" y="364" width="130" height="20" rx="4" />

Resumes agent run

<g transform="translate(400, 340)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="68" rx="10" />

  <text x="20" y="22">
    Agent
  </text>

  <text x="20" y="48">
    Adds sources
  </text>
</g>

<g transform="translate(400, 460)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="68" rx="10" />

  <text x="20" y="22">
    Agent
  </text>

  <text x="20" y="48">
    Refreshes report
  </text>
</g>

<g transform="translate(400, 580)" filter="url(#md-card-shadow-af)">
  <rect width="200" height="68" rx="10" />

  <text x="20" y="22">
    Outcome
  </text>

  <text x="20" y="48">
    Approved
  </text>
</g>

In responding to an interrupt, you delegate an [action](/reference/action) to the agent. Actions represent the decisions and modifications you make to a Middesk report. Available action types include:

| Type             | Description                                    |
| ---------------- | ---------------------------------------------- |
| `decision`       | Approve or reject the business                 |
| `dismissal`      | Dismiss a watchlist result as a false positive |
| `verify_tin`     | Mark a TIN as verified via document            |
| `add_sources`    | Add external data sources to the business      |
| `add_attributes` | Add addresses or people to the business        |

For the full action schema and type-specific payloads, see the [Action reference](/reference/action).

## Reviewing the result

After you respond to an interrupt, or an [interrupt policy](/work-with-agents/interrupt-policy) resolves it automatically, the agent carries out the resolution. Middesk records what the agent found, how each finding was resolved, which actions it carried out, and what changed on the Business report so you can [review the result](/work-with-agents/review-work) and check whether screenshot evidence is ready. When a resolution doesn't change the report, the result still records the outcome.

> **Get a demo**
>
> Contact your account manager or [contact sales](https://www.middesk.com/contact-sales) to inquire about access.

> Drive meaningful action based on agent findings and recommendations.