> ## Documentation Index
> Fetch the complete documentation index at: https://docs.woes.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent

> Planned agent, Lab, citation, confidence, and handoff categories for the Woes public API.

Agent APIs will help teams inspect and integrate grounded AI support behavior. The goal is visibility and workflow integration, not exposing hidden prompts or provider internals.

## Planned resources

| Resource               | Purpose                                                         |
| ---------------------- | --------------------------------------------------------------- |
| Lab runs               | Test questions and outputs used before customer launch.         |
| Agent drafts           | Suggested responses before operator approval.                   |
| Agent responses        | Customer-facing answers generated by the agent.                 |
| Citations              | Source evidence used to support an answer.                      |
| Confidence signals     | Indicators that a question was answerable or needed handoff.    |
| Clarification requests | Questions the agent asks before answering.                      |
| Handoff events         | Moments when automation stops and an operator should take over. |
| Verification summaries | Safe summaries of live API checks.                              |

## Common use cases

<CardGroup cols={2}>
  <Card title="Quality review" icon="chart-no-axes-combined">
    Track answer quality, citation coverage, and low-confidence themes.
  </Card>

  <Card title="Escalation workflow" icon="handshake">
    Route handoff events into internal support or engineering processes.
  </Card>

  <Card title="Source improvement" icon="book-open-check">
    Find questions where missing docs caused clarification or handoff.
  </Card>

  <Card title="Operator tooling" icon="bot">
    Connect drafts and citations to custom review experiences.
  </Card>
</CardGroup>

## Safety expectations

Agent APIs should:

* Avoid hidden prompts.
* Avoid provider routing internals.
* Redact secrets before returning or logging data.
* Prefer citations and confidence signals over unsupported answers.
* Preserve operator/customer boundaries.
* Hand off when context is missing or unsafe.

<Warning>
  Public APIs should not expose raw prompt chains, provider credentials, private notes, or internal-only debug traces.
</Warning>