Agent setup contract (interactive)

Use this contract when an agent adapts Ship to a real project.

Phase 1 — discovery interview (required)

The agent must ask and confirm:

  1. Tracker: system, states, labels/fields, API limitations.
  2. Scheduler/CI: what can run on cron/manual/webhook.
  3. Agent runtime: Cursor/Codex/Claude/Copilot/custom.
  4. Quality model: manual QA role, QA automation scope, regression cadence.
  5. Release policy: manual or scheduled prod promote, required gates.
  6. Communication: digest/retro email recipients (recommend DL aliases).
  7. Constraints: compliance, data residency, no-go actions.

Phase 2 — adaptation proposal

Agent provides 1-2 concrete implementation options with trade-offs:

  • fast/simple option,
  • stricter/governance-first option.

User chooses one before file edits.

Phase 3 — implementation

Agent creates/updates:

  • setup runbook,
  • mapping docs (tracker states/labels equivalents),
  • automation entrypoints,
  • quality/release gates,
  • daily digest + retro definitions.

Phase 4 — validation

Agent must show:

  • what was changed,
  • what still requires human secrets/permissions,
  • first green-path test plan,
  • rollback path.

Non-negotiable rules

  • No secret commits.
  • No destructive changes without explicit approval.
  • No silent assumptions when infrastructure is unknown.
  • Every automated transition must leave evidence.