nick-vels/skills

What it solves

Autopilot is a development framework designed for users who want to describe a project idea in plain language and receive a fully functional, "turnkey" software project without needing to read specifications, manage tasks, or write code. It eliminates the the common problem where AI agents lose track of original requirements during long development cycles by strictly tracking every requirement from brief to final delivery.

How it works

Autopilot transforms a user's brief into a structured development process across eight phases: preparation, requirements gathering, briefing, specification, planning, development, code review, and acceptance.

Key technical mechanisms include:

• Requirement Tracking: It breaks the brief into numbered requirements that are tracked throughout the project. A "blind acceptance" phase uses a separate agent to verify the final result against the original text, not the specification.
• Task-Based Development: It divides the project into "waves" of tasks. Each task is handled by a separate agent with a clean memory to prevent context pollution, with independent tasks running in parallel.
• Customizable Control: Users can choose between three modes (Full Auto, Semi-Auto, Manual) and three levels of depth (Strict, Normal, Deep) to control how much the agent asks and how much it infers.
• Project Memory: It automatically creates AGENTS.md or CLAUDE.md files to document the project's architecture and setup for future sessions.
• Visual Dashboard: It generates a self-updating HTML dashboard that tracks project progress, brief coverage, time spent, and technical debt in real-time.

Who it’s for

It is primarily for non-programmers or users who want a "just build it" experience with AI agents (like Claude Code, Cursor, or Codex) without getting involved in the line-by-line coding process.

Highlights

• Blind Acceptance: A separate agent verifies the final product against the original brief to ensure no requirements were dropped.
• Real-time Dashboard: An offline HTML file that tracks progress, timers, and task completion without needing a server.
• Crossover-Proof Memory: Automatic generation of project documentation for AI agents to prevent expensive re-learning in new sessions.
• Parallel Agent Execution: Independent tasks are processed by multiple agents simultaneously to reduce build time.
• Polish Mode: An optional phase that compares the final result against a provided reference (screenshot, site, or text) to iteratively refine the UI/UX.