Studio Core

Astra Studio

A Claude Code plugin marketplace for planning, designing, validating, and shipping plugins.

Astra Studio handles the outer loop of plugin development — business analysis, domain modeling, plugin validation, and promotion. Individual skill authoring and testing is executed through the official skill-creator as an internal pipeline capability.

Plugins

Plugin	Skills	What it does	Install
studio-core	4	Workspace management — init, promote, status, create-expert	claude plugin install studio-core@astra-studio
studio-insight	6	Business analysis toolkit — personas, journeys, processes, domains	claude plugin install studio-insight@astra-studio
studio-planner	5	Planning pipeline — event storming, domain modeling, skill design, build orchestration	claude plugin install studio-planner@astra-studio
studio-quality	2	Quality assurance — plugin validation, MCP wiring	claude plugin install studio-quality@astra-studio
studio-docs	6	Document delivery — blueprints, writing experts, parallel generation, export	claude plugin install studio-docs@astra-studio
studio-platform	6	Platform docs — brainmap, agent mapping, tech designs, visualization, speech	claude plugin install studio-platform@astra-studio

Quick Start

# 1. Register the marketplace
claude plugin marketplace add github:VanLengs/astra-studio-plugins

# 2. Install what you need
claude plugin install studio-core@astra-studio
claude plugin install studio-insight@astra-studio
claude plugin install studio-planner@astra-studio
claude plugin install studio-quality@astra-studio
claude plugin install studio-platform@astra-studio

# 3. Initialize studio in your project
/studio-core:init

# 4. Plan a plugin
/studio-planner:plan "your business domain"

# 5. Confirm the build plan
# → Astra Studio generates specs, then produces initial skill drafts in {target_dir}/

# 6. Test and iterate each skill with real inputs
# → Use skill-creator to refine individual skills

# 7. Validate the plugin
/studio-quality:validate {target_dir}

# 8. Ship a version (design docs stay active for next iteration)
/studio-core:promote {plugin-name}

studio-insight: Standalone Artifact Skills

These 6 skills produce professional deliverables independently — no pipeline required:

/studio-insight:persona-insight "全职妈妈，两个孩子"    # → persona card + empathy map
/studio-insight:journey-map "日常营养管理流程"           # → journey map + emotional curve
/studio-insight:process-flow "从记录饮食到生成周报"      # → process diagram + decision points
/studio-insight:domain-canvas "儿童健康管理"             # → domain boundary map + classification
/studio-insight:behavior-matrix "儿童健康管理"           # → actor × action × event × data table
/studio-insight:opportunity-brief "儿童健康管理"         # → prioritized opportunity assessment

studio-planner: Planning Pipeline

/studio-planner:plan chains 5 pipeline skills:

Step 1: event-storm
  Multi-role brainstorming (PM, architect, domain experts)
  → Invokes: persona-insight, journey-map, process-flow
  → Identifies: KB dependencies, expert scope (planning vs runtime)
  → Writes: studio/changes/{domain}/event-storm.md

        ↓ user confirms

Step 2: domain-model
  Clusters events into business domains, draws plugin boundaries
  → Full analysis mode: invokes domain-canvas, behavior-matrix, opportunity-brief
  → Fast mode: skips insight tools, goes directly to plugin proposals
  → Writes: studio/changes/{domain}/domain-map.md
  → Creates: {target_dir}/ scaffold for each plugin candidate

        ↓ user confirms

Step 3: skill-design
  Breaks each plugin into skills with interfaces and data flow
  → Detects plugin traits: stateful, hil-gated, kb-dependent, multi-pipeline, expert-scoped
  → Assesses complexity: prompt-only / scripts / MCP
  → Writes: studio/changes/{plugin}/skill-map.md (includes traits + pipelines)

        ↓ user confirms

Step 4: spec-generate
  Auto-generates specification files + trait-conditional scaffolding
  → Design docs → studio/changes/ (brief.md, plugin.json.draft)
  → Implementation → {target_dir}/ (SKILL.md skeletons, commands)
  → If stateful: init-workspace skill + runtime config/status templates
  → If hil-gated: approval gate sections in relevant skills
  → If multi-pipeline: per-pipeline orchestration commands
  → Advances status: planning → building

Step 5+: build, test, validate, promote
  build-skills produces initial skill drafts via skill-creator (working first drafts)
  → Test each skill with real inputs, iterate with skill-creator
  /studio-quality:validate validates {target_dir}/ and advances to approved
  /studio-core:promote creates a versioned milestone (v0.1 → v0.2)
  → Design docs are snapshotted to archive but stay active for next iteration

studio-platform: Platform Documentation Generator

After completing the planning pipeline, generate a full platform documentation suite:

# Generate all platform docs
/studio-platform:platform-docs {domain-name}

# Generate specific steps only
/studio-platform:platform-docs {domain-name} --steps brainmap,mapping

# With reference .pen file for visualization
/studio-platform:platform-docs {domain-name} --reference /path/to/ref.pen

Pipeline (6 steps):

Step 1: generate-brainmap
  → brainmap-index.md + brainmap-module-1~N.md
  Organizes agents by module with skill/data/model dependencies

Step 2: generate-agent-mapping
  → agent-plugin-mapping.md
  Maps agents ↔ plugins ↔ skills with architecture diagrams

Step 3: generate-tech-designs  (parallelized)
  → data-warehouse-design.md     (Kimball dimensional modeling)
  → data-collection-design.md    (IoT/edge-cloud pipelines)
  → knowledge-graph-design.md    (ontology, sub-graphs, reasoning)
  → ml-models-design.md          (model cards, MLOps pipeline)
  → rag-system-design.md         (knowledge bases, hybrid retrieval)

Step 4: generate-project-plan
  → project-plan.md              (Gantt, milestones, team, risks)

Step 5: generate-platform-visual
  → {platform-name}.pen          (4-frame design visualization)

Step 6: generate-speech
  → 迎检话术-{platform-name}.md   (presentation speech script)

Domain-agnostic — works for any industry vertical (elderly care, education, healthcare, etc.).

Subagent Roles

Both studio-insight and studio-planner use multiple perspectives via subagent roles:

Role	Perspective	Used by
Product Manager	User personas, journey mapping, prioritization	persona-insight, journey-map, opportunity-brief, event-storm
Architect	System boundaries, dependencies, feasibility	process-flow, domain-canvas, behavior-matrix, domain-model
Domain Expert	Domain-specific knowledge, real-world constraints	All 6 insight skills (dynamically discovered)

11 built-in domain experts ship with studio-insight — general roles (UX researcher, data analyst, compliance officer, operations manager) plus health and beauty domain experts.

Customizing Experts

# Create a new domain expert
/studio-core:create-expert 宠物营养专家

# Customize a built-in expert with your company's terminology
/studio-core:create-expert customize product-manager

Custom experts are saved to studio/agents/ (git-tracked, team-shared) and automatically discovered by all insight skills at runtime. Project-level experts override built-ins with the same filename.

Full Workflow

/studio-core:init                       /studio-planner:plan
        ↓                                       ↓
    studio/                             event-storm
    ├── config.yaml                       ├── persona-insight   → personas/
    ├── changes/                          ├── journey-map       → journeys/
    └── archive/                          ├── process-flow      → processes/
                                          ├── KB dependencies   → event-storm.md
                                          └── expert scope      → event-storm.md
                                                ↓
                                        domain-model (full / fast mode)
                                          ├── domain-canvas     → domain-canvas.md    ┐
                                          ├── behavior-matrix   → behavior-matrix.md  ├ full mode only
                                          └── opportunity-brief → opportunity-brief.md┘
                                                ↓
                                        skill-design
                                          ├── trait detection   → skill-map.md
                                          └── skill breakdown   → skill-map.md
                                                ↓
                                        spec-generate + trait scaffolding
                                                ↓
                                studio/changes/{plugin}/    {target_dir}/
                                ├── brief.md                skills/{skill}/SKILL.md
                                └── plugin.json.draft       commands/{skill}.md
                                                            commands/{pipeline}.md  (multi-pipeline)
                                                            skills/init-workspace/  (stateful)
                                                            agents/                 (expert-scoped)
                                                ↓
                              build-skills (initial drafts)
                                                ↓
                              test + iterate with skill-creator
                                                ↓
                              /studio-quality:validate
                                                ↓
                              /studio-core:promote (versioned: v0.1 → v0.2)
                                                ↓
      studio/archive/{plugin}/{date}-iteration-{N}/   {target_dir}/.claude-plugin/
      (snapshot — originals stay in changes/)          plugin.json (version bumped)

Architecture

Astra Studio is a marketplace (collection of plugins), not a monolithic plugin. Each plugin is independently installable:

• studio-core: Zero dependencies. Manages the studio/ workspace.
• studio-insight: Zero dependencies. Standalone business analysis tools. Useful beyond plugin development.
• studio-planner: Depends on studio-core and studio-insight. Orchestrates the planning pipeline.
• studio-quality: Zero dependencies. Can validate any plugin, not just studio-managed ones.
• studio-platform: Depends on studio-core. Generates industry AI model platform documentation from planning artifacts.

The studio/ directory is git-tracked — it holds development documentation (briefs, design decisions, status) with version control value. Inspired by OpenSpec's spec-driven workspace pattern.

Domains evolve incrementally — re-running the pipeline updates artifacts in-place (git diff = revision history), changelog.md logs each iteration, and only affected plugins (create/modify) get change workspaces.

Promotion creates versioned milestones — design docs are copied (not moved) to studio/archive/{plugin}/{date}-iteration-{N}/, so the active workspace remains available for continued iteration. Version numbers increment automatically (v0.1.0 → v0.1.1).

Development

# Test locally
claude --plugin-dir ./studio-core --plugin-dir ./studio-insight --plugin-dir ./studio-planner --plugin-dir ./studio-quality --plugin-dir ./studio-docs --plugin-dir ./studio-platform

License

Apache-2.0