For agentic companies

Assess what Humbleaf can do for your workflows

IAM tells you who can reach something. Humbleaf tells you whether this exact action should happen now. This path lets your operator delegate a bounded assessment to your own agent — then inspect the same artifacts and decide whether one guard integration is worth a pilot.

Operator: open consoleTry intent preview

Operator path

  1. 01Create an operator account at console.agentchain.cloud and open Discovery (pilot).
  2. 02Optional: run anonymous intent preview with your company’s coarse goal (workflow family + available input).
  3. 03Issue discovery access and copy the private invitation (objective, allowance, workspace, credential).
  4. 04Send your agent the invitation plus the assessor mission — no operator passwords.
  5. 05Review the returned session in your workspace; ask follow-ups; end access when the assignment is complete.

What the assessment establishes

  • Consequential operations in one operator-approved API document
  • Document-derived classifications kept separate from assessor-reported context
  • A route-agnostic proceed / revise / defer / no-responsible-recommendation result
  • Separate risk and pilot priorities, explicit unknowns, and proposed non-executing boundaries

What it does not establish

  • Prove savings, production readiness, or org-specific facts without your input
  • Replace an Authority Surface Map or legal/compliance sign-off
  • Execute refunds, payouts, or config changes on your systems
  • Access your workspace without discovery access issued by your operator

Assessor mission skeleton (customize; keep credentials out of public pages)

MISSION: HUMBLEAF WORKFLOW VALUE ASSESSMENT (YOUR COMPANY)

You are an assessor agent. Read https://docs.agentchain.cloud/agent-operators/discovery-pilot-start only for machine steps.

Company objective: [operator fills — e.g. which workflow first, what decision must be explicit]
Available input: [reference sample | approved OpenAPI/spec | none yet]
Exploration allowance: [operator declares or "unknown"]

Phase 1 — Anonymous intent preview (no credential). Preserve excludedExperimentIds. If the
first response asks candidateFocus, send a second preview with root-level previewAnswers.
Import the exact completed input; server recomputation fails closed on stale or missing data.

Phase 2 — Use discovery access from the operator invitation (X-Discovery-Grant). Create one
assessment owned by that grant. Follow only the selected track:
- fit_shortlist: return fit brief and missing inputs.
- document_only_review: submit one approved source, disposition questions, submit a structured
  recommendation, explicitly finish findings review, and return findings/recommendation/unknowns.
- refund/policy-comparison track: create a draft and run a synthetic comparison.

Phase 3 — Return separate risk and pilot rationales, provenance, proposed
ALLOW / APPROVAL_REQUIRED / DENY hypotheses, and honest limits. A recommendation,
comparison, or ALLOW result authorizes nothing. No operator JWT, production credential,
payment credential, or repo inspection.

Marketing context (not API contract): Humbleaf controls consequential actions before they execute and proves the decision after.

Machine contract: discovery pilot START. The START contract branches by selected track. Refund templates remain one optional family; findings-only review never requires a refund draft or comparison.

Begin

Need a human-led Authority Surface Map first?

The agent path is for teams already comfortable delegating a documented API or reference sample. Larger inventories start with /contact — no credentials required.

Map a workflow — no credentialsWhat's real today