shipkit-dev-team
Scannednpx machina-cli add skill stefan-stepzero/shipkit/shipkit-dev-team --openclawshipkit-dev-team - Framework Development Team Orchestrator
Purpose: Coordinate parallel framework development using Agent Teams
What it does:
- Reads a plan from
.claude/plans/{feature}.json - Creates a coordinated team tailored to framework development
- Assigns tasks with file ownership boundaries
- Sets up quality gates using local dev skills
- Monitors progress and handles phase transitions
When to Invoke
User says:
- "Team build this"
- "Create a dev team"
- "Parallel implement the plan"
- "Run the self-improvement team"
- "Team review this"
Automated trigger:
- After
/shipkit-dev-planproduces a plan - When the plan has 3+ parallel tasks (otherwise solo is fine)
Prerequisites
Required:
- Agent Teams enabled:
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 - A plan at
.claude/plans/{feature}.json(unless using a built-in template)
Helpful context:
.claude/specs/{feature}.json— spec for the featureCLAUDE.md— framework rules- Current git branch and status
Team Templates
Template: skill — New Skill Development
When: Building a new distributed skill (7-file integration).
Team:
| Role | Model | Responsibilities |
|---|---|---|
| Lead (you) | — | Coordinate, approve gates, run validation |
| Implementer A | Sonnet | Write SKILL.md + references |
| Implementer B | Sonnet | Update integration files (manifest, settings, routing, overview) |
| Reviewer | Opus | Validate against spec, check Skill Value Test, design quality |
File ownership:
- A:
install/skills/shipkit-{name}/ - B:
install/settings/,install/profiles/,install/skills/shipkit-master/,docs/generated/,install/claude-md/ - Reviewer: read-only everywhere
Phase flow:
- A writes SKILL.md → Gate: file exists, frontmatter valid
- B updates integration files → Gate: 7-file checklist passes
- Reviewer validates → Gate: review verdict PASS
- Lead runs
/shipkit-framework-integrity
Template: refactor — Multi-File Refactor
When: Structural changes across multiple framework components.
Team:
| Role | Model | Responsibilities |
|---|---|---|
| Lead (you) | — | Coordinate, resolve conflicts |
| Implementer per cluster | Sonnet | Own a file cluster (skills, agents, hooks, config) |
| Reviewer | Opus | Check nothing breaks, pattern consistency |
File ownership: Derived from plan's ownership clusters.
Template: review — Multi-Lens Review
When: Thorough review of framework changes from multiple perspectives.
Team:
| Role | Model | Responsibilities |
|---|---|---|
| Lead (you) | — | Synthesize findings |
| Design Reviewer | Opus | Skill Value Test, scope discipline, pattern consistency |
| Compatibility Reviewer | Sonnet | CC compatibility, deprecated patterns, frontmatter validity |
| Integration Reviewer | Sonnet | 7-file integration, cross-references, manifest sync |
All reviewers are read-only.
Template: release — Release Pipeline
When: Preparing a new Shipkit version.
Team:
| Role | Model | Responsibilities |
|---|---|---|
| Lead (you) | — | Version bump, changelog, final push |
| Validator | Sonnet | Run framework-integrity, validate changed skills |
| Auditor | Sonnet | Secrets scan, installer check, count audit |
Template: self-improve — Shipkit Self-Improvement
When: Running the full intelligence → implementation cycle.
Team:
| Role | Model | Responsibilities |
|---|---|---|
| Lead (you) | — | Coordinate pipeline, approve specs/plans |
| Scout | Haiku | Run /shipkit-scout → scout-report.json |
| Analyst | Sonnet | Run /shipkit-analyst → analyst-report.json |
| Ideator | Sonnet | Run /shipkit-ideator → opportunities.json |
Phase flow:
- Scout fetches CC docs and changelog → Gate: scout-report.json written
- Analyst maps against Shipkit → Gate: analyst-report.json written
- Ideator brainstorms → Gate: opportunities.json written
- Lead presents top opportunities to user for selection
- (Optional) Lead runs
/shipkit-dev-spec+/shipkit-dev-planfor selected opportunity - (Optional) Spawn implementation team from plan
Note: Phases 1-3 are serial (each reads previous output). Phase 5-6 can reuse the skill or refactor template.
Process
Step 1: Select Template or Load Plan
If --template specified, use the corresponding template above.
Otherwise, read .claude/plans/{feature}.json:
- Extract phases, tasks, ownership clusters, parallel groups
- Determine team size from task count and cluster count
Step 2: Write Team State
Create .claude/team-state.local.json:
{
"feature": "{feature-name}",
"template": "skill|refactor|review|release|self-improve",
"planPath": ".claude/plans/{feature}.json",
"specPath": ".claude/specs/{feature}.json",
"createdAt": "2026-02-20T...",
"teammates": {
"implementer-a": {"cluster": "A", "tasks": ["1.1", "1.2"]},
"implementer-b": {"cluster": "B", "tasks": ["2.1", "2.2"]},
"reviewer": {"cluster": "readonly", "tasks": ["3.1"]}
},
"phases": [
{"id": 1, "status": "in_progress", "gate": "..."},
{"id": 2, "status": "pending", "gate": "..."}
]
}
Step 3: Create Spawn Prompts
For each teammate, build a prompt that includes:
# You are {role} on the Shipkit framework development team
## Your Assignment
{tasks from plan, with specific file paths}
## File Ownership
You own: {list of file paths/directories}
DO NOT modify files outside your ownership boundary.
## Context
- Read CLAUDE.md for framework rules
- Read .claude/specs/{feature}.json for requirements
- Read .claude/plans/{feature}.json for the full plan
## Skills Available
{relevant local dev skills for this role}
## Quality Requirements
- All acceptance criteria from spec must be met
- Your changes must pass validation
- Report completion to lead when done
## Communication
- Message lead when you hit a blocker
- Message reviewer when ready for review
- Broadcast to team if you discover a shared concern
Step 4: Create the Team
Use Claude Code's team creation capabilities:
- Create team with name:
shipkit-{feature}-{date} - Create shared task list derived from plan
- Set task dependencies from plan
- Spawn teammates with their prompts
- Assign tasks to teammates
Step 5: Monitor and Gate
As lead:
- Watch task completions —
TaskCompletedhook validates build/test - Verify phase gates — When all tasks in a phase complete, verify the gate condition
- Unblock next phase — Mark gate task complete to unblock dependent tasks
- Handle blockers — If a teammate reports a blocker, help resolve or reassign
Step 6: Finalize
When all tasks complete:
- Run
/shipkit-dev-review --scope recent - If review passes, ask user if they want to release
- If yes, run
/shipkit-dev-release
When This Skill Integrates with Others
Before This Skill
/shipkit-dev-plan— Produces the plan this skill executes- Trigger: Plan must exist (unless using built-in template)
/shipkit-dev-spec— Produces the spec the plan implements
After This Skill
/shipkit-dev-review— Review completed team work/shipkit-dev-release— Release if review passes
During Execution (by teammates)
/shipkit-framework-integrity— Validation during and after implementation/shipkit-scout,/shipkit-analyst,/shipkit-ideator— Self-improvement template
Context Files This Skill Reads
.claude/plans/{feature}.json— Implementation plan (required unless template).claude/specs/{feature}.json— Feature specCLAUDE.md— Framework rules.claude/team-state.local.json— Previous team state (if resuming)
Context Files This Skill Writes
.claude/team-state.local.json— Active team state (for hooks)
Source
git clone https://github.com/stefan-stepzero/shipkit/blob/main/.claude/skills/shipkit-dev-team/SKILL.mdView on GitHub Overview
shipkit-dev-team coordinates parallel Shipkit framework development using Agent Teams. It reads a feature plan from .claude/plans/{feature}.json, builds a coordinated team, and assigns tasks with clear file ownership. It also sets up quality gates and monitors progress to manage phase transitions from development through review to release.
How This Skill Works
The workflow reads a plan at .claude/plans/{feature}.json and spins up a coordinated team tailored to framework development. It assigns tasks with explicit file ownership boundaries, applies local quality gates, and tracks progress through phase transitions. It supports templates for different scopes (skill, refactor, review, release, self-improve) and relies on Agent Teams (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1).
When to Use It
- Team build this
- Create a dev team
- Parallel implement the plan
- Run the self-improvement team
- Team review this
Quick Start
- Step 1: Ensure Agent Teams are enabled (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1) and a plan exists at .claude/plans/{feature}.json.
- Step 2: Pick a template that matches the scope (skill, refactor, review, release, or self-improve) and prepare the plan/specs as needed.
- Step 3: Run the shipkit-dev-team flow to spawn roles, assign ownership, and begin parallel work with progress monitoring.
Best Practices
- Enable Agent Teams before invoking (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1) and verify plan availability.
- Point to a plan at .claude/plans/{feature}.json or use a built-in template to scope work.
- Define clear file ownership boundaries for each role to prevent conflicts.
- Enforce gates early: ensure frontmatter validity, existence of required files, and the 7-file checklist where applicable.
- Choose the correct template (skill, refactor, review, release, self-improve) to align team roles with the feature scope.
Example Use Cases
- A new distributed skill integration uses the 'skill' template with Lead, Implementers, and a Reviewer, assigning SKILL.md and integration files to specific owners.
- A large multi-file refactor uses the 'refactor' template to cluster components (skills, agents, hooks, config) under coordinated ownership.
- A thorough multi-lens review uses the 'review' template with Lead, Design Reviewer, Compatibility Reviewer, and Integration Reviewer.
- Preparing a Shipkit version uses the 'release' template with Lead, Validator, and Auditor handling version bump and checks.
- A complete self-improvement cycle runs via the 'self-improve' template to optimize patterns and rules across the framework.