Scripted sample walkthrough · No real run data

Claude Code keeps working while your evening stays yours.

Watch a Claude Code run stay quiet when nothing needs you, surface a window-closing choice, queue a non-urgent ask, and return ready to resume.

Live walkthrough

Watch the run unfold in one place.

4:42 PM

Now playing: Claude Code keeps working while your evening stays yours.

Now playing

4:42 PM

Sample run via Claude sample_run_alpha

Nothing needs you

Claude Code keeps working while your evening stays yours.

Nothing needs you. Claude Code is working inside the approved path, using sample metadata only.

The demo starts calm: no prompts, code, paths, logs, repo names, or model output are sent to HeadsDown.

Nothing needs you Claude Code approved path metadata only

Product boundary

Claude/Pi controls the agent. HeadsDown controls the run.

Granting broad machine access can reduce tool approval prompts, but it does not decide when to interrupt, what should queue, or how to resume without rework.

Sample proof

This walkthrough uses deterministic sample IDs, sample timestamps, metadata buckets, and synthetic receipts only.

Product surface snapshots

Receipts tied to the live story

These compact sample cards show the command center, run detail, replay, and proof views without pushing the active demo below the fold.

Current state

Window closing

Extend to keep going, or wrap and save the handoff. If Sam does nothing, Claude Code keeps working.

Sample demo data 30m window Claude metadata only targeted validation

Sample demo data only. The action buttons use production action keys; the details button is illustrative. These controls do not mutate real runs. Sample boundary: window closing at 5:12 PM.

Sample run detail

Sample

Privacy-safe handoff metadata and approved actions for the same scripted run.

Approved bounds

Sam approved a small coding-agent run with one targeted validation step.

sample_run_alpha30m windowmetadata only

Handoff saved

The sample handoff is marked available for resume through privacy-safe metadata without exposing prompts, code, paths, or logs.

continuation artifactvalidation passed1 queued decision

Ready to resume

Morning review can resume approved work or keep the queued decision in place.

resume_runkeep_queued

Sample intervention replay

Receipt

A replay-style receipt explains what HeadsDown saw, the play it called, and the result.

  1. What was about to happen

    The run was expanding past the approved slice while validation was still pending.

  2. Scope drift caught

    The sample receipt keeps the run pointed at a user-elected decision, never an automatic stop.

  3. Result

    Run contained, handoff saved, and the non-urgent ask queued for morning.

Ready to resume

1

Confidence: sample

Sample handoff says the next approved step is ready.

Ask queued

1

Confidence: sample

Non-urgent decision waited for morning review.

Handoff saved

1

Confidence: sample

Ready to resume without starting over.

Prompts/code/paths/logs stored

0

Confidence: sample

Queries in, decisions out, never content.

Ready to let HeadsDown keep your agent runs moving?

Create an account to define your rules, then watch how supported agent integrations stay quiet, surface review, queue for morning, wrap, or resume without guessing when to interrupt.

Set up your first rule See the full product story