ai-delivery-spec
AI Delivery Spec is a requirement management skill for human and AI collaboration that turns an idea, existing system, or change request into a single baseline that business, engineering, testing, and coding agents can use. It solves the problem of missing metrics, states, permissions, failures, and acceptance rules discovered after implementation starts. It is for product managers, engineers, testers, and coding agents working on consumer, enterprise, or government products.
✨ Key features
- Routes to the lightest suitable path and stopping point automatically.
- Closes key unknowns through strategic, systemic, observable-behavior, and devil's-advocate lenses.
- Produces one unified PRD with a human-readable body and engineering appendix.
- Generates operable HTML prototypes in product mode, review mode on request.
- Provides deterministic local gates distinguishing BLOCK, P0_UNKNOWN, GAP, PASS.
- Uses stable IDs and machine contracts for coding agent execution.
🎯 Use cases
- Clarify a vague product idea into a deliverable requirement baseline.
- Turn existing materials or a legacy system into a reviewable PRD.
- Generate an HTML prototype for customer demos or requirement confirmation.
- Manage a change request by comparing against a baseline across roles and data.
- Formalize acceptance criteria into execution records with evidence.
📦 Installation
🧰 Requirements: Python 3.10+ with PyYAML and jsonschema for local gates; no Python needed for conversational use. Works with Agent Skills-compatible tools like Codex, Claude Code, Cursor, Trae, and OpenClaw.
npx skills add franklinxkk/ai-delivery-specopenclaw skills install @franklinxkk/ai-delivery-spec🚀 Usage
/ads 我有一个“企业数据一键上报”的想法，请判断目前最需要澄清什么，并带我做到可交付。
/ads I have an idea for one-click enterprise data reporting. Identify the most important unknown and guide me to a deliverable result.
❓ FAQ
Do I need to learn internal stages or templates to use this?
No. You can start with /ads and the Skill automatically routes to the lightest suitable path and stopping point based on your goal.
Can I use this with Spec Kit or OpenSpec?
Yes, you can hand off through stable requirement IDs and acceptance evidence, but AI Delivery Spec does not require these tools and does not claim built-in export or integration.
What does the gate check do?
It runs deterministic local checks that classify outputs as BLOCK, P0_UNKNOWN, GAP, or PASS, and it retains not_proven status to avoid false confidence.
Is Python required to use the Skill?
No, Python is only needed for running the deterministic local gates. Conversational use works without it.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.