Skip to main content
Proof in ADE is intentional evidence — a screenshot, recording, browser trace, console log, or verification note — captured at the moment a result is visible to show that a change actually does what it claims. Agents often say they tested something; proof makes that claim inspectable, and ties it to the work it belongs to. Nothing watches your agent and files screenshots for it. When the agent (or you) reaches a checkpoint worth showing — a login succeeds, a form submits, an error reproduces, a test passes — proof is captured on purpose and normalized into one canonical artifact model that a reviewer can skim in under a minute.

Evidence, not chatter

Concrete artifacts a reviewer can open, instead of a claim buried in a transcript.

Tied to the work

Every proof links to a chat, lane, PR, or Linear issue — and can link to several at once.

Intentional capture

Captures are deliberate. There’s no auto-observer; the agent or you decide a moment deserves evidence.

​
Proof kinds

ADE normalizes every capture into one of five canonical kinds, whatever tool produced it.
KindUse it for
ScreenshotShow the visible state of an app or browser.
RecordingCapture a short interaction or repro.
Browser tracePreserve browser steps and diagnostics from a test tool.
Console logsAttach useful runtime or browser output.
Verification noteRecord what was checked and why it passed.
Inspecting a page in ADE's browser pane

​
What proof attaches to

Every artifact carries at least one owner. A single artifact can link to more than one, so it surfaces wherever the work is reviewed — and evidence can flow from an exploratory chat to a PR comment without losing provenance.
  • Chat — the proof drawer below the composer in the session that produced it.
  • Lane — alongside the worktree’s diff and review.
  • PR — surfaced at pull request closeout, where reviewers look.
  • Linear issue — attached to the issue when Linear is connected.
  • Automation run — captured as part of a dispatched agent or worker run.

​
Where it fits

1

An agent reaches a checkpoint

A UI change renders, a flow completes, a test passes, an error reproduces.
2

Proof is captured

The agent or you capture a screenshot, recording, trace, log, or note with a short caption.
3

A reviewer sees it

The artifact rides along with the chat, lane, PR, or issue — so the result is verifiable, not just asserted.
Capture proof at the moment the result is visible, and write captions like a teammate is reading them. A good proof set is three to eight captures with captions a reviewer can read in one pass — not a screenshot after every click.

Capture proof

Prerequisites and the ade proof capture commands.

Review proof

The capture, review, and attach lifecycle in detail.