team-ship

name	team-ship
description	Build and deploy features — branch, implement, commit, review, test, deploy, PR/merge
argument-hint	<feature description>
disable-model-invocation	true
allowed-tools	Bash(git *), TaskCreate, TaskUpdate, TaskList, TaskGet

Ship — Build and Deploy Features

Implement a new feature or enhancement from start to production.

Feature to implement: $ARGUMENTS

Execution

With Agent Teams (recommended)

Enable with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

Create a team and spawn teammates:

1. Create team named ship-<feature-slug>
2. Define acceptance criteria before implementation begins:
   • Break the feature into testable behaviors and success conditions
   • Write these criteria into claude-progress.json so all agents share the same definition of "done"
   • This prevents self-evaluation bias where agents confidently approve their own mediocre work
3. Spawn teammates: shipper, full-stack-developer, database-admin (if data changes needed), reviewer, documentor

Create the following task list with dependencies:

#	Task	Owner	Blocked By
1	Create feature branch feature/<name> from main	shipper	—
2	Implement the feature end-to-end	full-stack-developer	1
3	Implement database/data layer changes (if needed)	database-admin	1
4	Commit all changes with conventional message	shipper	2, 3
5	Review implementation for security, bugs, performance	reviewer	4
6	Fix review findings (only if reviewer returns REQUEST_CHANGES)	full-stack-developer + database-admin	5
7	Run full test suite	shipper	5 (verdict APPROVE or APPROVE_WITH_WARNINGS)
8	Fix regressions (if tests fail — loop back to 7)	full-stack-developer	7
9	Update documentation for feature changes	documentor	7
10	Deploy and create PR to main	shipper	7, 9

Parallelism:

• Tasks 2 and 3 run simultaneously after task 1.
• If task 6 fires, application and data fixes can run simultaneously.

Loops:

• Review gate: If task 5 returns REQUEST_CHANGES, create task 6 (fix findings) and re-run task 5. Repeat until the verdict is APPROVE or APPROVE_WITH_WARNINGS. Tasks 7+ stay blocked until the gate clears.
• Test gate: If task 7 finds failures, create task 8 (fix regressions) and re-run task 7.

Without Agent Teams (fallback)

Execute sequentially using the Task tool:

1. Task(shipper, "Create feature branch feature/<name> from main")
2. Task(full-stack-developer, "Implement: $ARGUMENTS")
3. Task(database-admin, "Implement data layer changes for: $ARGUMENTS") — only if needed
4. Task(shipper, "Commit all changes with message: feat: <description>")
5. Task(reviewer, "Review the implementation on this branch")
6. If reviewer returns REQUEST_CHANGES: Task(full-stack-developer, "Fix review findings: <findings>") (and Task(database-admin, ...) if data-layer findings), then re-run step 5. Loop until APPROVE or APPROVE_WITH_WARNINGS.
7. Task(shipper, "Run full test suite")
8. If tests fail: Task(full-stack-developer, "Fix test failures: <failure details>"), then re-run step 7
9. Task(documentor, "Update documentation for: $ARGUMENTS")
10. Task(shipper, "Deploy and create PR to main")

Workflow Diagram

┌─────────┐   ┌──────────────────┐   ┌─────────┐   ┌──────────┐   ┌─────────┐   ┌────────────┐   ┌──────────┐
│ Shipper │──►│ Full-Stack Dev + │──►│ Shipper │──►│ Reviewer │──►│ Shipper │──►│ Documentor │──►│ Shipper  │
│ Branch  │   │  DB Admin  (║)   │   │ Commit  │   │  Review  │   │  Test   │   │ Update Docs│   │Deploy+PR │
└─────────┘   └──────────────────┘   └─────────┘   └────┬─────┘   └────┬────┘   └────────────┘   └──────────┘
                                                       │ ▲             │ ▲
                                                       ▼ │             ▼ │
                                          [REQUEST_CHANGES]     [test failures]
                                          Dev fixes findings    Dev fixes regressions
                                          → re-review           → re-run tests

  Tests stay blocked until reviewer returns APPROVE or APPROVE_WITH_WARNINGS.