Your first task and result

Continue your original intention after setup. Replace the bracketed placeholders. Copying a request does not execute it or start a trial.

Change existing code — free source help

Make [requested change] in [existing file]. Use StatePort to highlight consequential choices I have not specified, with references to the source you changed.

The integration begins a review of existing source before editing. A useful result names a consequential choice, its consequence and a checked current source anchor. The interpretation remains agent-reported; StatePort checks the reference, not the agent’s motives or completeness. New files without a baseline, inaccessible workspaces and stale anchors cannot produce that receipt. None of these limitations should block ordinary coding work.

For an already completed diff, ask for a read-only source review. It can give relevant code links but cannot create a pre-edit baseline retroactively. Source lookup alone is not a source-choice observation. A typo or no-op may correctly produce no extra note.

Check the running app — Developer access

Use StatePort to open my app at [local URL] and check [expected behaviour]. Tell me what you observed and what remains unverified.

Expect the actual actions, criterion, target/conditions, outcome and unrun limitations. Source analysis is not browser evidence. Live-only work is not saved replay. Access requires explicit acceptance when needed.

Save a difficult case — Free Capture

Capture this case from my app at [local URL]. Let me review the retained material before saving it. Do not replay it or start a trial.

Expect mandatory Review before Save, a Card ID, what it retains and a way to return. Saved does not mean verified. In a new chat ask the agent to find that Card and inspect its revision. For Node human Open use npx stateport open <state-card-id>; this opens the current Card, not a pinned historical revision. Check identity before comparison. Developer execution denial preserves the Card.

Read the result

Keep four things clear: what happened, its source/evidence, what remains unverified, and how to inspect or repeat it. A comparable fix claim needs the same Card/revision, target and criteria before and after a change. Outcomes may pass, fail or remain inconclusive; human acceptance is separate.

An optional introduction is prepared at most once per local runtime profile. Preparation is not confirmed display in the host. Ask explicitly to disable the introduction or optional chat observations; these preferences are independent and persist across restarts. Required evidence limitations remain visible.