Customer Research Agent

This is an example recipe, not a core Little Harness concept. Use it when an agent needs internal tools, large intermediate datasets, and a clean export boundary.

Shape

const harness = createHarness({
  host: localHost(),
  model,
  system: "Prepare evidence-backed customer research for account teams.",
  tools: {
    queryCustomers,
    searchDocs,
    searchIncidents,
    getUsageMetrics,
    saveReport,
  },
  skills: [skill("./skills/customer-renewal-analysis")],
});

Working Pattern

The agent should:

1. Query account metadata.
2. Fetch recent incidents and usage metrics.
3. Write large datasets to artifacts.
4. Analyze those artifacts with scripts or shell tools.
5. Produce a report artifact.
6. Call an export tool only when the report should become app data.

Persistent Context

Use a Persistent Dir for account-team preferences, not for every raw customer record.

persistentDirs: [
  {
    harnessDir: "/persistent/account-team",
    load: ({ extraBody }) => accountMemory.loadFolder(extraBody.teamId),
    store: ({ extraBody, changes }) =>
      accountMemory.applyChanges(extraBody.teamId, changes),
  },
]

Keep raw source-of-truth data behind tools so permissions and freshness remain under app control.

Export Tools

Useful exports include:

• save_report
• create_crm_note
• draft_email
• create_followup_task

Make the side effect obvious in the tool name and description.