Skip to main content

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.

  1. Pick a kit, then install: pipeline-kit init --ide cursor (kit mode) or pipeline-kit init --mode orchestrator --ide cursor. See Two kits.
  2. Paste the AGENTS.md routing table and a short product blurb. See Adapt AGENTS.md.
  3. Edit .pipeline/config.json:
    • verify.rules for your folders (or leave [])
    • deploy.targets for your apps
    • intake.jira.enabled (false unless you use Jira)
  4. Rewrite the local-deploy runbook and deploy-local.sh for your stack. See Adapt local deploy.
  5. Skim .pipeline/docs/DOCUMENT-STANDARD.md before editing any other pack markdown.
  6. Add features/ and .pipeline/state/ to .gitignore.
  7. Open the repo root in the IDE. Kit mode: confirm the run-workflow skill is visible. Orchestrator mode: confirm CURSOR_API_KEY and try pipeline-kit run --dry-run.
  8. Smoke test. Kit mode: ask “How does X work?” (expect ask) and “Add a small label change” (expect feature-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.