A skills.sh compatible skill for AI agents to converge fuzzy requirements into verifiable specs before writing code.
RDD is a development methodology for AI Agent software engineering. Its core goal is not to make LLMs generate code faster, but to build a complete requirements convergence chain from human intent to reliable software implementation:
Human Intent
↓
Requirement Engineering (What to build? Why? Where are the boundaries?)
↓
Specification Engineering (What behavior must the system exhibit?)
↓
Implementation (Code + tests, AC as test contracts)
↓
Verification (Run real app, browser automation, integrity checks)
# Install via skills.sh CLI
npx skills add anyJohn/OpenReq
# Or clone manually
git clone https://github.com/anyJohn/OpenReq.git ~/.claude/skills/rdd- Work Items: RR → Feature → Story → Task, with status tracking
- Spec: Structured behavior/constraint/invariant model with uncertainty labels (confirmed/inferred/unresolved)
- Decisions: First-class artifacts, frozen after human approval
- Three scenario paths: A (incremental), B (full replacement), C (lightweight)
- Artifact shelf-life: permanent / migration-only / session-only
- AC as test contract: No separate TEST files — tests written with implementation
- Lite channel: Skip US split for small changes, single-file spec
- Git as artifact store: Commit message + docs/specs/ for solo dev
OpenReq/
├── skills/
│ └── rdd/
│ ├── SKILL.md # Main skill (frontmatter + full methodology)
│ └── references/
│ ├── templates.md # RR / Spec / Traceability templates
│ └── rdd-design.md # Full design document (30+ sections)
├── skills.sh.json # Category metadata for skills.sh
├── README.md
└── LICENSE
Once installed, tell your AI agent:
用 RDD 流程开发：把 web 从 Vue 迁移到 React
RDD: 需求收敛一下，这个功能还没想清楚
先理清再动手，别直接改代码
The agent will follow the RDD flow: Impact Analysis → Requirement Engineering → Specification → Implementation → Verification.
MIT