New-project checklist
Use this on every new engagement. The full customer surface is AGENTS.md + config.json, then deploy and test files if your runners are not the defaults.
- Pick a kit, then install:
pipeline-kit init --ide cursor(kit mode) orpipeline-kit init --mode orchestrator --ide cursor. See Two kits. - Paste the
AGENTS.mdrouting table and a short product blurb. See Adapt AGENTS.md. - Edit
.pipeline/config.json:verify.rulesfor your folders (or leave[])deploy.targetsfor your appsintake.jira.enabled(falseunless you use Jira)
- Rewrite the local-deploy runbook and
deploy-local.shfor your stack. See Adapt local deploy. - Skim
.pipeline/docs/DOCUMENT-STANDARD.mdbefore editing any other pack markdown. - Add
features/and.pipeline/state/to.gitignore. - Open the repo root in the IDE. Kit mode: confirm the
run-workflowskill is visible. Orchestrator mode: confirmCURSOR_API_KEYand trypipeline-kit run --dry-run. - Smoke test. Kit mode: ask “How does X work?” (expect
ask) and “Add a small label change” (expectfeature-development). Orchestrator mode:pipeline-kit run --slug try --workflow ask --request "How does X work?" --runner fake.
Optional later: tracker MCP, wiki pages after retro, knowledge, memory bank (uv tool install -e ".[memory]" then memory link), plugins and observability.