claudeAutoPilot — Summary

claudeAutoPilot is a PRD-driven autonomous development framework for Claude Code that transforms a Product Requirements Document into a managed Kanban workflow with automated quality gates (security scan, code review, test), environment setup, and deployment lifecycle management.

Problem it solves: Claude Code is a powerful assistant but requires the developer to manually manage the development lifecycle; claudeAutoPilot imposes a 4-column kanban (Backlog → In Progress → QA → Live) driven by PRD tasks, automating the transitions between stages and surfacing only the decisions that truly require human input.

Distinctive trait: The kanban board is a generated HTML file (docs/kanban/kanban_dev.html) — no external project management tool required — and the framework's decision loop routes only specific decision types (production deploy, architecture change, scope change, security issue) to the human, handling everything else autonomously.

Target audience: Solo developers and small teams using Claude Code who want to turn a PRD into a shipping application with minimal workflow overhead; the framework installs as a .claude/commands/ + .autopilot/ folder pair dropped into any project.

Production-readiness: v3.1.0 commands; MIT license; maintained by Travis Sutphin with 0 stars and 1 contributor; last pushed February 2026. Small project, early-stage.

Differs from seeds: Closest to agent-os in its Markdown-scaffold + bash-script architecture (no MCP, no skills beyond commands), but claudeAutoPilot adds an automated kanban board generation, PRD-driven task extraction with a Node.js validator, a 4-stage DevOps workflow with explicit QA gates, and daily digest/end-of-day commands. Unlike BMAD-METHOD (multi-persona orchestration), claudeAutoPilot is single-agent with human-mediated decision gates. Unlike taskmaster-ai (task decomposition via MCP), claudeAutoPilot uses bash scripts and simple JSON config files with no external server.

claudeAutoPilot — Origin and Philosophy

Origin

Created by Travis Sutphin (github.com/travissutphin) as an open-source project under MIT license. Repository: travissutphin/openSource-claudeAutoPilot. The author also maintains rapidPRD.app for AI-assisted PRD creation, suggesting the framework is part of a larger PRD-to-production toolchain.

Philosophy (from README)

"Turn a PRD into a production app with AI handling the process. You make decisions. AI does everything else."

On the decision model:

"Claude handles 90% of workflow automatically. You're only asked when:

• Production deploy: 'Task #15 ready for production. Approve?'
• Architecture change: 'Should we use Redis or in-memory caching?'
• Scope change: 'This feature needs 2 more tasks. Proceed?'
• Security issue: 'Medium vulnerability found. Fix now or defer?'"

On the workflow:

"Break down the PRD into tasks, Create and manage a kanban board, Move tasks through a 4-column DevOps workflow, Run tests, security scans, and code reviews automatically, Deploy to staging, verify, and request production approval, Only notify you when decisions are needed."

Design principles

1. PRD-first — the framework's entry point is always a PRD document; no PRD → no workflow
2. Kanban as artifact — project state lives in an HTML kanban board committed to the repo
3. Decision taxonomy — decisions are classified (production, architecture, scope, security) and only the relevant decision types surface to the human
4. Daily cadence — [StartDay] / [EndDay] commands structure the workday around task execution

claudeAutoPilot — Architecture

Distribution

• Type: standalone-repo (clone-and-use)
• CLI binary: none
• Source: https://github.com/travissutphin/openSource-claudeAutoPilot
• Version analyzed: v3.1.0 (command version)
• License: MIT

Install

git clone https://github.com/travissutphin/openSource-claudeAutoPilot.git my-project
cd my-project
# Place PRD at docs/prd/PRD.md
# Open Claude Code: claude
# Run: [SetupProject]

Required runtime

• Claude Code
• Node.js 18+ (for automation scripts)
• Git

Target AI tools

Claude Code exclusively.

Directory structure

my-project/
├── .autopilot/
│   ├── automation/          # Node.js scripts (10 scripts)
│   │   ├── ai-code-review.js
│   │   ├── evaluate-progression.js
│   │   ├── file-generator.js
│   │   ├── generate-secrets.js
│   │   ├── iterative-refinement.js
│   │   ├── kanban-updater.js
│   │   ├── pattern-analyzer.js
│   │   ├── populate-env.js
│   │   ├── prd-validator.js
│   │   └── setup-database.js
│   ├── config/              # Configuration JSON files
│   │   ├── approval-levels.json
│   │   ├── decision-taxonomy.json
│   │   ├── environments.json
│   │   ├── placeholders.json
│   │   ├── quality-gates.json
│   │   ├── refinement-rules.json
│   │   └── workflow-states.json
│   ├── docs/                # Framework documentation
│   │   ├── SETUP-NEW-PROJECT.md
│   │   └── TASK-LIFECYCLE.md
│   ├── examples/            # Sample PRDs
│   │   └── sample-prd-saas.md
│   └── templates/           # Templates for generated files
├── .claude/
│   └── commands/            # 12 slash commands
├── docs/
│   ├── prd/
│   │   └── PRD.md           # YOUR PRD GOES HERE
│   └── kanban/
│       └── kanban_dev.html  # Generated Kanban board
├── CLAUDE.md                # Generated AI instructions
└── .env.example             # Generated environment template

Workflow states (4-column kanban)

claudeAutoPilot — Components

Slash Commands (12)

Node.js Automation Scripts (10)

Config files (7)

Generated artifacts

• CLAUDE.md — AI instructions generated from PRD and project config
• docs/kanban/kanban_dev.html — Interactive HTML kanban board
• .env.example — Environment template from PRD tech requirements
• .autopilot/docs/decisions/pending.md — Pending decisions queue

claudeAutoPilot — Uniqueness and Positioning

Differs from seeds

claudeAutoPilot is closest to agent-os in its Markdown-scaffold + bash-script architecture (no MCP server, no skills beyond commands), but adds a PRD-driven setup phase, an HTML kanban board, a decision taxonomy classification system, and a daily cadence structure (StartDay/EndDay). Unlike taskmaster-ai which uses an MCP server and JSON task files, claudeAutoPilot uses Node.js automation scripts and a generated HTML kanban board. Unlike BMAD-METHOD which coordinates multiple named personas across the full development lifecycle, claudeAutoPilot is single-agent with a simpler 4-stage kanban. The decision-only notification model (routing only 4 decision types to the human) is closer to a "lights-out operations" philosophy than any seed framework explicitly implements.

Positioning

claudeAutoPilot targets developers who want to ship a PRD to production without managing a separate project management tool. The generated HTML kanban board is the project management surface, and Claude Code handles the workflow execution. The framework is explicitly "Claude Code only" — no cross-tool portability.

Observable failure modes

1. PRD quality bottleneck — if the PRD is thin or vague, task extraction fails and the entire framework has nothing to work from; prd-validator.js adds a quality gate but can't fix a bad PRD
2. HTML kanban limitation — a static HTML file doesn't support real-time multi-session collaboration; only one Claude session should update the board at a time
3. Small project maturity — 0 stars, 1 contributor, last push Feb 2026; the framework is early-stage and may have rough edges
4. No cross-tool portability — entirely Claude Code-specific; can't be used with Gemini CLI, Cursor, etc.
5. Decision routing reliability — the "only 4 decision types surface to human" promise depends on Claude correctly classifying all decisions; misclassification could lead to autonomous decisions that should have involved the human

Explicit antipatterns

• Starting without a PRD
• Running [SetupProject] multiple times without archiving the old kanban
• Manually editing the kanban HTML (use the automation scripts instead)