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:
- Tracker: system, states, labels/fields, API limitations.
- Scheduler/CI: what can run on cron/manual/webhook.
- Agent runtime: Cursor/Codex/Claude/Copilot/custom.
- Quality model: manual QA role, QA automation scope, regression cadence.
- Release policy: manual or scheduled prod promote, required gates.
- Communication: digest/retro email recipients (recommend DL aliases).
- 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.