OpenCode Product Pipeline — Orchestrator

Overview

This skill coordinates a 7-step multi-agent pipeline that transforms a raw customer idea into a fully planned, implementation-ready project. Each step is handled by a specialized skill that produces structured markdown outputs.

Command / Invocation

This is the explicit entry point for the pipeline. Invoke it by:

• Slash command (OpenCode / compatible tools): /start-new
• Natural language: type start-new or "start new product pipeline"
• Idea-driven: describe a product idea and the skill will offer to begin the pipeline

When invoked, it initializes state and starts at Step 1 (Interview Agent). Use "Continue the product pipeline" to resume an existing run from pipeline-outputs/<project-slug>/pipeline-state.json.

Pipeline Steps

State Tracking

Each project gets its own isolated directory. Maintain a pipeline-state.json inside pipeline-outputs/<project-slug>/ with this structure:

{
  "current_step": 1,
  "completed_steps": [],
  "project_name": "",
  "project_slug": "",
  "idea_summary": "",
  "output_dir": "pipeline-outputs/<project-slug>",
  "tech_stack": {
    "frontend": "",
    "backend": "",
    "database": "",
    "infrastructure": "",
    "mobile": "",
    "testing": "",
    "ci_cd": "",
    "other_tools": "",
    "constraints": ""
  },
  "tech_preferences_confirmed": false,
  "outputs": {
    "01-interview-brief.md": "",
    "02-business-analysis.md": "",
    "03-product-requirements.md": "",
    "04-ux-design.md": "",
    "05-architecture.md": "",
    "06-sprint-plan.md": "",
    "07-implementation-plan.md": ""
  },
  "notes": []
}

Workflow

1. Initialize: Ask the user for a project name. Derive a project_slug (lowercase, hyphenated, no spaces). Create pipeline-outputs/<project-slug>/ directory and pipeline-outputs/<project-slug>/pipeline-state.json.
2. Execute Step-by-Step: For each step:
   • Load the corresponding skill
   • Pass the accumulated context from all previous steps, including the output_dir value (pipeline-outputs/<project-slug>)
   • Each skill replaces its @output_dir@ placeholder with this value
   • Save output to pipeline-outputs/<project-slug>/<output-file>
   • Update pipeline-outputs/<project-slug>/pipeline-state.json
   • Confirm completion with the user before proceeding
3. Technology Preferences Checkpoint (after Step 4, before Step 5):
   • Ask the user explicitly about their preferred tech stack:
     • Frontend framework/library (React, Vue, Svelte, Angular, etc.)
     • Backend language/framework (Node.js, Python/Django, Go, Rust, Java/Spring, etc.)
     • Database (PostgreSQL, MySQL, MongoDB, SQLite, etc.)
     • Infrastructure / hosting (cloud provider, containers, serverless, etc.)
     • Mobile platform (if applicable: React Native, Flutter, Swift/Kotlin, etc.)
     • Testing approach (framework preferences, coverage expectations)
     • CI/CD pipeline (GitHub Actions, GitLab CI, Jenkins, etc.)
     • Any constraints (team expertise, existing systems, licensing, cost limits)
   • Save the confirmed preferences to pipeline-state.json under tech_stack
   • Set tech_preferences_confirmed: true
4. Handle Iterations: If a step needs revision, update the output and state, then re-confirm
5. Resume Support: If pipeline-outputs/<project-slug>/pipeline-state.json exists, read it and resume from current_step

Rules

• NEVER skip a step without explicit user approval
• ALWAYS save outputs as markdown files before proceeding
• ALWAYS update pipeline-outputs/<project-slug>/pipeline-state.json after each step
• Each project gets its own pipeline-outputs/<project-slug>/ directory — never share or overwrite across projects
• If the user says "continue", read state and resume
• If the user says "go back to step X", update state and re-run that step
• Ask clarifying questions if the user's idea is ambiguous
• Each step's output becomes the input context for the next step
• Use the project name/slug when referring to the output directory in each skill call
• Technology preferences must be confirmed with the user before Step 5 — do not skip or assume
• Pass the confirmed tech_stack from pipeline-state.json to the solution architect and developer agent as part of their context