{
  "id": "page:library-pilot-shell",
  "_kind": "Page",
  "_file": "wiki/library/pilot-shell.md",
  "_cluster": "wiki",
  "attributes": {
    "nodeKind": "Page",
    "title": "Pilot Shell Methodology for Babysitter SDK (Library)",
    "displayName": "Pilot Shell Methodology for Babysitter SDK (Library)",
    "slug": "library/pilot-shell",
    "articlePath": "wiki/library/pilot-shell.md",
    "article": "\n# Pilot Shell Methodology for Babysitter SDK\n\nAdapted from [Pilot Shell](https://github.com/maxritter/pilot-shell) by Max Ritter -- a spec-driven development framework with strict TDD, quality hooks, and persistent memory.\n\n## Overview\n\nPilot Shell is a comprehensive development methodology that enforces spec-driven planning, strict test-driven development (RED->GREEN->REFACTOR), automated quality pipelines, and cross-session persistent memory. This adaptation implements all core workflows as Babysitter SDK process definitions with full orchestration support.\n\n## Process Files\n\n| File | Process ID | Description | Tasks |\n|------|-----------|-------------|-------|\n| `pilot-shell-orchestrator.js` | `pilot-shell/orchestrator` | Full lifecycle: mode routing -> plan -> implement -> verify -> merge | 17 |\n| `pilot-shell-feature.js` | `pilot-shell/feature` | Feature mode: semantic search -> spec -> TDD -> unified review -> merge | 14 |\n| `pilot-shell-bugfix.js` | `pilot-shell/bugfix` | Bugfix mode: analysis -> behavior contract -> test-before-fix -> verify | 9 |\n| `pilot-shell-sync.js` | `pilot-shell/sync` | Codebase sync: explore -> index -> conventions -> rules | 8 |\n| `pilot-shell-quality-pipeline.js` | `pilot-shell/quality-pipeline` | Quality hooks: lint/format/typecheck -> TDD -> context monitoring | 9 |\n\n## Skills\n\n| Directory | Name | Description |\n|-----------|------|-------------|\n| `skills/spec-driven-development/` | spec-driven-development | Spec creation via semantic search and structured planning |\n| `skills/strict-tdd/` | strict-tdd | RED->GREEN->REFACTOR enforcement with compliance scoring |\n| `skills/behavior-contract/` | behavior-contract | Bug condition/postcondition formalization |\n| `skills/quality-hooks/` | quality-hooks | Language-specific lint/format/typecheck pipeline |\n| `skills/context-preservation/` | context-preservation | State capture/restore across compactions |\n| `skills/persistent-memory/` | persistent-memory | Cross-session observation storage and retrieval |\n| `skills/worktree-isolation/` | worktree-isolation | Git worktree management for safe development |\n| `skills/codebase-sync/` | codebase-sync | Convention discovery and rule generation |\n\n## Agents\n\n| Directory | Name | Role |\n|-----------|------|------|\n| `agents/plan-reviewer/` | plan-reviewer | Spec completeness validation, assumption challenging |\n| `agents/unified-reviewer/` | unified-reviewer | Compliance + quality + goal alignment review |\n| `agents/tdd-enforcer/` | tdd-enforcer | Test-first implementation verification |\n| `agents/file-checker/` | file-checker | Language-specific lint/format/typecheck |\n| `agents/context-monitor/` | context-monitor | Context usage tracking and threshold alerts |\n| `agents/spec-guard/` | spec-guard | Prevents premature completion of incomplete specs |\n| `agents/memory-curator/` | memory-curator | Observation capture and skill extraction |\n\n## Execution Modes\n\n### Feature Mode (`mode: 'feature'`)\nFull spec-driven lifecycle:\n1. **PLAN**: Semantic search -> clarifying questions -> spec -> plan-reviewer validation (conditional) -> approval\n2. **IMPLEMENT**: Git worktree -> strict TDD RED->GREEN->REFACTOR per task -> quality hooks -> full test suite\n3. **VERIFY**: Full tests -> unified review (compliance + quality + goals) -> auto-fixes -> squash merge\n\n### Bugfix Mode (`mode: 'bugfix'`)\nSystematic bug resolution:\n1. **ANALYSIS**: Trace to file:line -> root cause -> blast radius assessment\n2. **CONTRACT**: Bug Condition + Postcondition + Invariants = Behavior Contract\n3. **TEST-BEFORE-FIX**: Failing bug test (RED) -> preservation tests -> minimal fix (GREEN) -> verify\n4. **VERIFY**: Behavior Contract audit + full suite + lint\n\n### Quick Mode (`mode: 'quick'`)\nChat-based iteration without planning scaffolding. TDD still enforced.\n\n## Quality Pipeline\n\nSupported tool chains:\n- **Python**: ruff + pyright\n- **TypeScript/JavaScript**: prettier + eslint + tsc\n- **Go**: gofmt + golangci-lint\n\nAuto-fix convergence loop: up to 3 iterations of fix-and-recheck.\n\n## Model Routing (Informational)\n\nThe original Pilot Shell uses model routing. In Babysitter SDK, this is informational context for agent configuration:\n- **Planning**: Opus (reasoning-heavy tasks)\n- **Plan Verification**: Sonnet (conditional review)\n- **Implementation**: Sonnet (fast execution)\n- **Code Verification**: Sonnet (unified review)\n\n## Examples\n\nSee `examples/` directory for input JSON examples for each process.\n\n## Attribution\n\nAll processes, skills, and agents in this directory are adapted from [Pilot Shell](https://github.com/maxritter/pilot-shell) by Max Ritter. See `references.md` for detailed attribution.\n",
    "documents": [
      "specialization:pilot-shell"
    ]
  },
  "outgoingEdges": [
    {
      "from": "page:library-pilot-shell",
      "to": "specialization:pilot-shell",
      "kind": "documents"
    }
  ],
  "incomingEdges": [
    {
      "from": "page:index",
      "to": "page:library-pilot-shell",
      "kind": "contains_page"
    }
  ]
}