{
  "id": "page:library-gsd",
  "_kind": "Page",
  "_file": "wiki/library/gsd.md",
  "_cluster": "wiki",
  "attributes": {
    "nodeKind": "Page",
    "title": "GSD-Adapted Workflows for Babysitter SDK (Library)",
    "displayName": "GSD-Adapted Workflows for Babysitter SDK (Library)",
    "slug": "library/gsd",
    "articlePath": "wiki/library/gsd.md",
    "article": "\n# GSD-Adapted Workflows for Babysitter SDK\n\nThis directory contains workflows adapted from the [Get Shit Done](https://github.com/glittercowboy/get-shit-done) methodology for the Babysitter SDK orchestration framework.\n\n## Overview\n\nThese processes implement the GSD philosophy of systematic planning, research, execution, and verification while using the Babysitter SDK's task orchestration capabilities.\n\n## Core Workflows\n\n### 1. New Project Initialization (`new-project.js`)\n- Systematic vision capture through agent-guided questions\n- Parallel domain research (stack, features, architecture, pitfalls)\n- Requirements scoping (v1/v2 separation)\n- Roadmap creation with phased milestones\n\n### 2. Phase Discussion (`discuss-phase.js`)\n- Captures implementation preferences before planning\n- Identifies decision points (layouts, API patterns, structure)\n- Creates context document for guided research\n\n### 3. Phase Planning (`plan-phase.js`)\n- Agent-based research informed by discussion context\n- Creates atomic task plans with verification\n- Plan checker validates against requirements\n- Iterative refinement until plans pass quality gates\n\n### 4. Phase Execution (`execute-phase.js`)\n- Parallel task execution with fresh agent contexts\n- Atomic git commits per task\n- Progress tracking and dependency management\n- Automated verification gates\n\n### 5. Work Verification (`verify-work.js`)\n- User acceptance testing gates\n- Automated issue diagnosis\n- Fix plan generation\n- Re-verification loop\n\n### 6. Milestone Audit (`audit-milestone.js`)\n- Comprehensive milestone review\n- Definition-of-done verification\n- Quality assessment\n- Readiness determination\n\n## Advanced Workflows\n\n### 7. Codebase Mapping (`map-codebase.js`)\n- Brownfield project analysis\n- Architecture understanding\n- Pattern identification\n- Integration planning\n\n### 8. Iterative Convergence (`iterative-convergence.js`)\n- Quality-gated development loop\n- TDD with agent scoring\n- Refinement cycles until target quality reached\n- Parallel quality checks\n\n## Utilities\n\n### 9. State Management (`state-manager.js`)\n- Cross-session memory\n- Decision tracking\n- Blocker management\n- Progress persistence\n\n### 10. Research Orchestrator (`research.js`)\n- Parallel domain investigation\n- Stack analysis\n- Best practice discovery\n- Pitfall identification\n\n## Usage\n\nEach workflow is a self-contained process module that can be:\n\n1. **Run directly:**\n```bash\nbabysitter run:create \\\n  --process-id gsd/new-project \\\n  --entry gsd/new-project.js#process \\\n  --inputs inputs.json\n```\n\n2. **Composed into larger workflows:**\n```javascript\nimport { process as discussPhase } from './discuss-phase.js';\nimport { process as planPhase } from './plan-phase.js';\n\nexport async function process(inputs, ctx) {\n  const discussion = await ctx.task(discussPhase, inputs);\n  const plan = await ctx.task(planPhase, { ...inputs, discussion });\n  // ...\n}\n```\n\n3. **Customized for specific needs:**\nExtend or modify any workflow to match your project requirements.\n\n## Key Principles\n\n### Context Engineering\n- Always load relevant project context (PROJECT.md, REQUIREMENTS.md, STATE.md)\n- Keep agent contexts fresh with targeted information\n- Use artifacts directory for persistent state\n\n### Quality Gates\n- Verification loops at planning stage\n- Automated checks during execution\n- User acceptance gates at phase completion\n\n### Atomic Execution\n- Each task produces discrete, verifiable output\n- Git commits per task for bisect-ability\n- Clear success/failure criteria\n\n### Iterative Refinement\n- Agent-based quality scoring\n- Feedback loops until target quality reached\n- Progress tracking across iterations\n\n## File Organization\n\n```\ngsd/\n├── README.md                    # This file\n├── new-project.js              # Project initialization\n├── discuss-phase.js            # Phase discussion\n├── plan-phase.js               # Phase planning with verification\n├── execute-phase.js            # Parallel execution with commits\n├── verify-work.js              # UAT and automated diagnosis\n├── audit-milestone.js          # Milestone completion check\n├── map-codebase.js             # Brownfield analysis\n├── iterative-convergence.js    # Quality convergence loop\n├── state-manager.js            # State persistence utilities\n├── research.js                 # Research orchestration\n└── examples/                   # Example input files\n    ├── new-project-example.json\n    ├── phase-example.json\n    └── milestone-example.json\n```\n\n## Integration with Babysitter SDK\n\nThese workflows leverage:\n- **Agent tasks** for LLM-based reasoning and decision making\n- **Breakpoints** for human approval gates\n- **Parallel execution** via `ctx.parallel.all()`\n- **Task composition** for modular workflow building\n- **State persistence** via artifacts and journal\n- **Quality scoring** through agent-based assessment\n\n## Differences from Original GSD\n\n1. **SDK Native**: Uses Babysitter SDK primitives instead of custom orchestration\n2. **Language**: JavaScript/Node.js instead of prompt-based skills\n3. **Execution Model**: Task-based orchestration instead of CLI commands\n4. **State Management**: Journal and artifacts instead of markdown files\n5. **Composability**: Modular processes can be imported and composed\n\n## Getting Started\n\n1. Install the Babysitter SDK:\n```bash\nnpm i -g @a5c-ai/babysitter-sdk@latest\n```\n\n2. Initialize a new project:\n```bash\nbabysitter run:create \\\n  --process-id gsd/new-project \\\n  --entry gsd/new-project.js#process \\\n  --inputs '{\"projectName\": \"My App\"}'\n```\n\n3. Follow the workflow through discussion → planning → execution → verification\n\n## Contributing\n\nWhen adding new workflows:\n- Follow the GSD methodology of research → plan → verify\n- Use agent tasks for reasoning, node/shell tasks for execution\n- Include breakpoints at key decision points\n- Document expected inputs and outputs\n- Provide example input files\n",
    "documents": [
      "specialization:gsd"
    ]
  },
  "outgoingEdges": [
    {
      "from": "page:library-gsd",
      "to": "specialization:gsd",
      "kind": "documents"
    }
  ],
  "incomingEdges": [
    {
      "from": "page:index",
      "to": "page:library-gsd",
      "kind": "contains_page"
    }
  ]
}