First task walkthrough

Quick Start gets the tool installed. This page covers the next step: run one small, reviewable task and end with a pull request receipt.

This written walkthrough is the canonical first-task path: run a real small change, verify it, and leave a review receipt before merging.

Target outcome

Use a tiny issue: add a health check endpoint, fix one failing test, or update a doc link. The task should be small enough for a junior developer to review without trusting the agent blindly.

  • A real repository, not a mocked transcript.
  • A branch with one narrow change.
  • Tests, lint, or an explicit note explaining why none apply.
  • A pull request with the command run and the verification result.

Commands

npm install -g @0dai-dev/cli
cd your-project
0dai init
0dai doctor
0dai swarm run "add a /health endpoint and test it"

What to watch for

The useful part of the walkthrough is not that an agent typed code. It is the control loop around the code: the agent reads the project layer, scopes the task, writes a small patch, verifies it, and leaves a receipt.

MomentGood signStop if
0dai initai/ and native agent configs are created.The command wants secrets, production credentials, or broad repo access.
0dai doctorIt reports the project layer and agent config health.It cannot find the repo root or generated config files.
First taskThe branch touches only files needed for the issue.The agent rewrites unrelated files or expands the goal.
Pull requestThe PR body names the issue, files touched, and checks run.The PR claims a test passed without a command or log.

Review receipt

A good first PR body is short and specific. It should give a reviewer enough context to decide whether the change is narrow.

Fixes #123

Changed:
- added GET /health
- added route test

Verified:
- npm test -- health
- npm run lint

Video publish gate

The landing-page video should publish only after these are true:

  • It shows a real repository and a real pull request.
  • It includes faces and voices or a spoken walkthrough, not only a silent screencast.
  • Captions are present and accurate.
  • No metric, time, or customer claim is invented for the recording.
  • The final cut stays under 10 minutes and links back to this page.

Next up