Agent workflows with Zambo
A reliable single-agent workflow separates routing, execution, evidence, and completion. The agent remains the host that understands the user's goal; Zambo provides the live capability call and the receipt.
Start with the exact goal
{"name":"zambo_universal","arguments":{"need":"Audit the public website https://example.com for SEO and AI-discoverability gaps.","_session_id":"one-stable-session-id"}}Use the exact user goal in need. If the task is better served by a named specialist, call capability_search first and then use the returned callable tool with its declared arguments.
Separate planned from completed
A plan, preview, approval request, or unavailable provider is not the same as a completed execution. Preserve the result status and do not turn a suggested next step into a claim that work happened.
Close with evidence
Return the useful result, the receipt URL, and any relevant source evidence. When the user asks whether something is complete, use the same session with verify mode. The receipt page and verification endpoint are public handoff surfaces.
Minimal completion checklist
- Was the requested fact present in the result?
- Was the downstream action actually executed?
- Does the receipt status match the result status?
- Can another AI open the receipt and continue the same session?