Sign-off and approval flow

Business Summary: Sign-off is how you release accountable outcomes from governed agents — approving drafts, connector writes, or high-spend runs so the right people explicitly own each decision before it ships.

Governed agents never release sensitive outputs or execute external writes without the right human approval. Sign-off is your explicit action at a checkpoint — approve, request changes, or reject.

A decision commit means the AI's recommendation has been reviewed and locked by a human approver — the outcome is released to Outputs, a connector, or your operating memory.

How to approve a checkpoint

Open Overview and check Awaiting Approval — or open Process Board and find a run row with an HITL badge.
Click the pending item to open the workstream workspace.
Locate the checkpoint card in chat or the collaboration rail — look for Needs you, Needs your approval, or Sign-off.
Read the context: draft output, proposed connector write, or NTU impact.
Choose Approve, Request changes, or Reject.
After approval, confirm the checkpoint clears on Overview and the run resumes or publishes.

For tier definitions (Soft, Hard, Critical), see Human-in-the-loop (HITL) checkpoints.

Sign-off sits at the center of the HITL flow. After you confirm a Plan run, the agent executes until it hits a gate. Nimbus assigns the checkpoint to approvers on the workstream team; your sign-off either resumes the run, publishes an output, or executes an approved write.

Before execution — Confirm & run (approves the run plan, not a checkpoint):

sequenceDiagram
  participant O as Operator
  participant N as Nimbus
  participant S as Agent
  O->>N: Confirm and run with NTU quote
  N->>S: Start governed Plan run

During execution — checkpoint sign-off:

sequenceDiagram
  participant S as Agent
  participant N as Nimbus
  participant A as Approver
  S->>N: Reaches checkpoint
  N->>A: Needs you / Sign-off card
  A->>N: Approve, request changes, or reject
  alt Approved
    N->>S: Resume or commit decision
  else Rejected or changes
    N->>S: Revise or stop
  end

Where approvals appear

When a checkpoint is open for you, look for Needs you, Needs your approval, or Sign-off action buttons on the checkpoint card. These surface in workstream chat as HITL interrupt cards, Overview Pending approvals, workstream Checkpoints panels, and milestones when an approval gate blocks progress.

What you'll see: A checkpoint card with context (draft output, proposed write, spend amount), actions to approve or request changes, and a link to the workstream and run.

Checkpoint tiers

Tier	When it applies	Blocks execution?
Soft HITL	Clarification, missing data, low confidence	Usually no — surfaced in chat
Hard HITL	Draft report review before publish	Yes — until you sign off
Critical HITL	External connector writes, dual approvers, high spend	Yes — may require two approvers

For preset configuration (Standard, Strict, Custom), see Checkpoints and approvals.

Approval gate states

stateDiagram-v2
  [*] --> NotRequired
  NotRequired --> NeedsYou: HITL enabled
  NeedsYou --> Approved: Sign-off approve
  NeedsYou --> Changes: Request changes
  Changes --> NeedsYou: Revised draft
  NeedsYou --> Rejected: Reject
  Approved --> Committed: Decision commit
  Committed --> [*]
  Rejected --> [*]

The sign-off flow

The agent reaches a gate — output review, connector write, or spend threshold. Nimbus assigns the checkpoint to approvers on the workstream team. You review context on the card — draft, proposed action, NTU impact. You choose Approve, Request changes, or Reject. On approve, the run resumes or the output publishes — a decision commit. Activity records the approval, actor, and timestamp. Rejecting or requesting changes sends feedback to the agent for revision.

Before the run: Confirm & run

Plan mode runs also require confirmation before execution. The Confirm & run card is not the same as output sign-off — it approves the run plan (objective, data sources, estimated NTU, optional HITL toggles). After you confirm, checkpoints during execution still apply based on agent governance and your choices.

See .

Who can sign off

Typically workstream team members with approver role, workstream owner for Needs you items, a second approver when Strict governance requires dual sign-off, and org admins may override depending on the governance tier your admin has set — see Org access tiers. Guests and viewers cannot approve Critical writes.

Example scenario

RevOps runs a Plan mode agent to refresh pipeline forecasts. At Confirm & run, the FP&A lead approves the run plan and NTU quote. Mid-run, the agent pauses at Hard HITL with a draft board report. The workstream owner signs off; the report lands in Outputs. A follow-up run proposes CRM stage updates — Critical HITL assigns both the owner and a second approver before any write executes.

What to do next

Configure checkpoint tiers → Checkpoints and approvals
See where approved outcomes land → Where decisions land
Run your first governed Plan run → Running an agent

FAQs {#faqs}

Q: What's the difference between Confirm & run and sign-off? A: Confirm & run approves starting a governed Plan run. Sign-off approves a specific checkpoint during or after execution — output publish or connector write.

Q: I approved in chat but Overview still shows pending — why? A: Refresh or wait for sync. Ensure you completed Sign-off on the correct checkpoint, not just dismissed the card.

Q: Do Analyze mode charts need approval? A: Usually not unless governance requires output review for all artifacts. Formal reports and connector writes do.

Q: Can I disable all approvals? A: You can relax HITL in settings, but organizations with compliance needs should keep output review and external write gates enabled. See Human-in-the-loop.

Q: Where does the committed decision go after I approve? A: See Where decisions land.

Q: What happens if I request changes instead of rejecting? A: The agent receives your feedback and attempts a revision. The checkpoint stays open until you approve the updated draft or reject the run.