Skip to main content

Review pending approvals on the Dashboard

This guide shows you how to resolve every pending human task from the Dashboard's Pending Approvals tab — including approving, dismissing with feedback, and editing the tool's parameters before approving.

Use this workflow when:

  • You've been away and want to clear the backlog in one pass.
  • You run several AI Agents and don't want to open each chat to check.
  • You want to tweak the inputs before letting a tool run.

Open the Pending Approvals tab​

  1. Open the Dashboard from the sidebar.
  2. Click the Pending Approvals tab.

If you have no pending tasks, you'll see "All clear — Your agents are running autonomously. Nothing needs your attention."

If you do have pending tasks, each shows up as a card.

Anatomy of an approval card​

Each card shows:

  • The AI Agent (bold).
  • The tool the AI Agent wants to run (shown as a code-style label).
  • A short action description summarising what's about to happen (e.g. "Publish article: "How AI Agents Qualify Leads at Scale"").
  • Optional badges:
    • URGENT — see When a task is marked URGENT below.
    • ⚡ Blocking N parent agents — approving this will unblock the parent AI Agents that are waiting for the result.
  • A timestamp showing when the task was created (hover for the absolute time).
  • A Show parameters ▾ toggle that expands to show the exact inputs the tool will receive, key by key. Click Hide ▴ to collapse again.

Four action buttons sit at the end of each card:

ButtonWhat it does
✓ ApproveApproves and executes the tool immediately.
✎ EditOpens the task detail modal so you can edit the parameters before approving.
✕ RejectOpens the dismiss modal so you can send feedback and reject.
💬 ChatOpens the conversation this task belongs to.

Approve as-is​

Click ✓.

The tool runs with the parameters you saw. The card updates to:

✓ Approved — moved to execution queue {AI Agent} · {tool}

The AI Agent picks up where it left off. You can click the 💬 button on the resolved card to jump into the conversation and watch the tool's output land.

danger

Approving a destructive action (send email, CRM write, webhook call, delete data) runs immediately and cannot be undone. Use ✎ to review parameters if you're unsure.

Edit parameters before approving​

Click ✎.

A task detail modal opens with two tabs:

Parameters tab. Shows the action description, any chain badge (e.g. "⚡ Blocking 4 parent agents"), and an editable form with one field per parameter. Change any value you need.

Chat tab. Shows the conversation this task belongs to, so you can see what your AI Agent was thinking when it chose these parameters.

When you're ready:

  • Click ✓ Approve with Changes to approve with your edited values.
  • Click Cancel to close without approving — the task stays pending.

This is how you keep an AI Agent productive while still correcting small mistakes. You don't have to dismiss and make the AI Agent try again from scratch.

Dismiss with feedback​

Click ✕.

The dismiss modal opens with:

  • The action description.
  • An info note: "After dismissal, the agent will receive your feedback and continue the conversation. You can monitor its response in the Conversations tab."
  • A Feedback for the agent textarea with the placeholder "Tell the agent why and what to do instead…".

Type your feedback (optional but strongly recommended), then:

  • Click ✕ Dismiss with Feedback to confirm.
  • Click Cancel to back out.

The card updates to:

✕ Dismissed — feedback sent to agent {AI Agent} · {tool}

Your AI Agent reads the feedback and adjusts in the same conversation. Click 💬 on the dismissed card to see how it responded.

tip

Specific, actionable feedback beats a short "no". "Wrong URL — use the production domain, not staging" lets the AI Agent retry correctly. A blank dismiss just stops the task with no guidance.

When a task is marked URGENT​

A pending task shows the URGENT badge when both of these are true:

  1. It's in waiting_for_human — actively blocked on you.
  2. At least one of the following:
    • It's blocking 3 or more ancestor tasks (i.e. 3+ parent AI Agents or batches are parked in waiting_for_children because of this one task).
    • It's been waiting more than 24 hours since it was created.

In practice, that means URGENT tasks are the ones where your delay is costing the most:

  • A deeply-nested subagent approval is blocking a whole AI Team run.
  • A task has sat all day and nobody acted on it.

Non-urgent pending tasks are still pending — they just aren't yet at a level where the delay is expensive. Urgency is informational; the approval actions are the same. Clear URGENT items first when you have a backlog.

Jump into the conversation​

Click 💬 on any card (pending or resolved) to open the conversation this task belongs to. The same conversation view also shows the approval card inline, in context, so you can read what led up to the action.

Resolve questions cards​

Questions cards (from the Ask User Questions tool) also show up in the Pending Approvals tab. They open a wizard where you answer one question at a time and click Submit. See Answer AI Agent questions.

If a card looks weird​

SymptomLikely causeFix
Card doesn't respond to clicksThe task was resolved elsewhere (another tab, another team member, the chat).Refresh — it drops off the list.
💬 button doesn't open the conversationThe underlying conversation was archived.Open the AI Agent's Chat tab directly.
New pending tasks don't show up liveYour browser tab lost the WebSocket connection.Refresh the page.

For deeper diagnosis, see Troubleshoot stuck tasks.