Structured actions
Stable action names, explicit inputs, predictable outputs, and typed errors keep tool use understandable.
For agents / the contract
Tools that say
what happened.
SynthCrew is designed around the questions an agent needs answered: what will happen, why is it allowed, what changed, and what needs approval?
Talk through an integration ↗actioncampaign.previewstatus: ready_for_review
nexthuman approvalrequired: yes
01 / Agent primitives
Make the system
easy to query.
Dry-run before apply
Separate planning from side effects. Let an agent show affected records, warnings, and an evidence ID before applying.
Idempotent operations
Retries should not create duplicate campaigns, sends, approvals, or next actions.
State you can inspect
Expose operation state, blockers, approvals, and receipts without requiring a browser screenshot.
Human approval at the edge
Keep irreversible or externally visible actions behind policy and approval boundaries.
Evidence receipts
Return a verifier-readable record of meaningful mutations and their authorization context.
02 / A useful response
Not just
“success: true.”
operationcampaign.preview
statusneeds_review
affected42 leads · 0 sends
evidencereceipt_id returned
Build with the boundary