Back to the product tour

The walkthrough, in words

Inside CuttleCode.

The tour is a silent recording of the real CuttleCode interface in a dedicated Atlas Commerce example workspace. Desktop and mobile follow the same chapters; switching screens keeps your place.

0:00Project context

Give every session the project context.

Explore Atlas Commerce: a storefront, payments API, and design system with shared instructions, a review skill, and a scoped agent.

  1. Atlas Commerce groups the storefront, payments API and design system in one project. Filter the repository list, then open the shared project instructions. Money handling, API validation and keyboard access are part of the starting context for every session.

  2. Skills & agents sit directly below repositories. Open the checkout-review skill to read its checklist, then inspect the checkout-reviewer agent. The phone uses dedicated sheets for the same saved configuration.

0:22Context harness

See exactly what reaches the agent.

Open the context harness to see the instructions, skills, and agents that shape a session, along with their context footprint.

  1. Open the context harness to inspect its components and context layers. This makes the sources of the agent’s instructions visible before work starts.

  2. Read the composed context, including project instructions, repository guidance and selected skills. Shared rules live at the project level; repository-specific details stay beside their code.

0:33Intents & autoplay

Start with an intent. Keep it moving.

Read the generated plan, arm autoplay, and follow the implementation. Ask the Intent chat for status and an independent QA review. Long agent waits are cut out.

  1. Create “Make checkout keyboard-accessible” with automatic research enabled. A real agent reads the three repositories and identifies a click-only checkout control.

  2. A labeled cut skips the research wait. Read the resulting overview and implementation steps, then arm autoplay. It launches implementation in an isolated worktree. Open the work session to see its progress and its Intent group beside other real conversations.

  3. Return to Intent chat and ask for status and an independent QA session. The agent uses the Cuttle CLI to inspect the work and hand off the review.

  4. Inspect the resulting changes and review evidence. This is actual execution in a disposable checkout example; long waits are removed, and neither automated tests nor a passing review stand in for human approval.

1:26Cuttle CLI

Find past work. Set up the next run.

Ask an agent to find a previous review, follow its session link, and inspect the disabled weekday QA schedule it created. The scheduling request and wait are skipped. The agent uses the same Cuttle CLI you can script yourself.

  1. Ask a project assistant to find the previous payment review. It searches actual sessions using the Cuttle CLI and replies with a link.

  2. Follow the link and continue that conversation. Ask it to schedule weekday QA at 9am using defaults, keeping the automation disabled. The desktop request finds the existing matching pipeline schedule; the phone request creates a separate QA automation. Inspect the actual saved workflow.

1:44Pipelines

Two reviews. One combined report.

Watch two reviews execute in parallel, pass their gate, and finish a combined report. Waiting runs at 12×; milestones return to normal speed. Then inspect the completed timeline, metrics, and captured output variables.

  1. Watch the live Checkout review graph with payment-contract and keyboard-access reviews running in parallel. Their activity indicators stay animated.

  2. The waiting periods play at 12×. Playback slows to normal speed as each review completes, the all-success gate releases the summary task, and the entire run succeeds. All four nodes finish; successful review execution does not mean the example has no defects.

  3. The graph stays on screen through completion, with a brief normal-speed hold showing all four successful stages. Open Timeline and filter its gate event, inspect the Metrics summary, then read the four captured node outputs in Variables. This review produced no decision artefacts.

2:44Automations

Turn an event into a finished review.

Create a webhook automation, save its bearer key, and send a real request. See authentication reject a missing key, then follow the accepted event into a running agent and its completed checkout review. Waiting is accelerated; the demo key is revoked after recording.

  1. Create Checkout webhook review, select the project and reviewer, and declare the required change parameter. Save the generated bearer key and inspect the webhook authentication settings.

  2. Use the clearly labeled external HTTP client to send a real request. A missing key returns HTTP 401. Add the generated bearer key and send again; HTTP 202 accepts the event and creates an execution.

  3. Follow execution history into the agent session. Watch it inspect the checkout repositories, run the storefront tests and return its review. Waiting is accelerated; milestones return to normal speed. Each device records its own real run. Read the completed findings and remaining unverified checks in the saved conversation. The displayed demo keys are revoked after capture. The tour then fades to a CuttleCode splash and back to its opening before looping.