specialization:pilot-shell
Pilot Shell reference
Adapted 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.
Pilot Shell Methodology for Babysitter SDK
Adapted from Pilot Shell by Max Ritter -- a spec-driven development framework with strict TDD, quality hooks, and persistent memory.
Overview
Pilot 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.
Process Files
| File | Process ID | Description | Tasks |
|---|---|---|---|
pilot-shell-orchestrator.js | pilot-shell/orchestrator | Full lifecycle: mode routing -> plan -> implement -> verify -> merge | 17 |
pilot-shell-feature.js | pilot-shell/feature | Feature mode: semantic search -> spec -> TDD -> unified review -> merge | 14 |
pilot-shell-bugfix.js | pilot-shell/bugfix | Bugfix mode: analysis -> behavior contract -> test-before-fix -> verify | 9 |
pilot-shell-sync.js | pilot-shell/sync | Codebase sync: explore -> index -> conventions -> rules | 8 |
pilot-shell-quality-pipeline.js | pilot-shell/quality-pipeline | Quality hooks: lint/format/typecheck -> TDD -> context monitoring | 9 |
Skills
| Directory | Name | Description |
|---|---|---|
skills/spec-driven-development/ | spec-driven-development | Spec creation via semantic search and structured planning |
skills/strict-tdd/ | strict-tdd | RED->GREEN->REFACTOR enforcement with compliance scoring |
skills/behavior-contract/ | behavior-contract | Bug condition/postcondition formalization |
skills/quality-hooks/ | quality-hooks | Language-specific lint/format/typecheck pipeline |
skills/context-preservation/ | context-preservation | State capture/restore across compactions |
skills/persistent-memory/ | persistent-memory | Cross-session observation storage and retrieval |
skills/worktree-isolation/ | worktree-isolation | Git worktree management for safe development |
skills/codebase-sync/ | codebase-sync | Convention discovery and rule generation |
Agents
| Directory | Name | Role |
|---|---|---|
agents/plan-reviewer/ | plan-reviewer | Spec completeness validation, assumption challenging |
agents/unified-reviewer/ | unified-reviewer | Compliance + quality + goal alignment review |
agents/tdd-enforcer/ | tdd-enforcer | Test-first implementation verification |
agents/file-checker/ | file-checker | Language-specific lint/format/typecheck |
agents/context-monitor/ | context-monitor | Context usage tracking and threshold alerts |
agents/spec-guard/ | spec-guard | Prevents premature completion of incomplete specs |
agents/memory-curator/ | memory-curator | Observation capture and skill extraction |
Execution Modes
Feature Mode (`mode: 'feature'`)
Full spec-driven lifecycle: 1. **PLAN**: Semantic search -> clarifying questions -> spec -> plan-reviewer validation (conditional) -> approval 2. **IMPLEMENT**: Git worktree -> strict TDD RED->GREEN->REFACTOR per task -> quality hooks -> full test suite 3. **VERIFY**: Full tests -> unified review (compliance + quality + goals) -> auto-fixes -> squash merge
Bugfix Mode (`mode: 'bugfix'`)
Systematic bug resolution: 1. **ANALYSIS**: Trace to file:line -> root cause -> blast radius assessment 2. **CONTRACT**: Bug Condition + Postcondition + Invariants = Behavior Contract 3. **TEST-BEFORE-FIX**: Failing bug test (RED) -> preservation tests -> minimal fix (GREEN) -> verify 4. **VERIFY**: Behavior Contract audit + full suite + lint
Quick Mode (`mode: 'quick'`)
Chat-based iteration without planning scaffolding. TDD still enforced.
Quality Pipeline
Supported tool chains:
- **Python**: ruff + pyright
- **TypeScript/JavaScript**: prettier + eslint + tsc
- **Go**: gofmt + golangci-lint
Auto-fix convergence loop: up to 3 iterations of fix-and-recheck.
Model Routing (Informational)
The original Pilot Shell uses model routing. In Babysitter SDK, this is informational context for agent configuration:
- **Planning**: Opus (reasoning-heavy tasks)
- **Plan Verification**: Sonnet (conditional review)
- **Implementation**: Sonnet (fast execution)
- **Code Verification**: Sonnet (unified review)
Examples
See examples/ directory for input JSON examples for each process.
Attribution
All processes, skills, and agents in this directory are adapted from Pilot Shell by Max Ritter. See references.md for detailed attribution.