惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

小众软件
小众软件
博客园 - Franky
罗磊的独立博客
G
Google Developers Blog
The GitHub Blog
The GitHub Blog
P
Proofpoint News Feed
Recent Announcements
Recent Announcements
V
V2EX
F
Fortinet All Blogs
阮一峰的网络日志
阮一峰的网络日志
Blog — PlanetScale
Blog — PlanetScale
月光博客
月光博客
U
Unit 42
GbyAI
GbyAI
A
About on SuperTechFans
WordPress大学
WordPress大学
Engineering at Meta
Engineering at Meta
雷峰网
雷峰网
Microsoft Azure Blog
Microsoft Azure Blog
Martin Fowler
Martin Fowler
D
DataBreaches.Net
The Cloudflare Blog
H
Hackread – Cybersecurity News, Data Breaches, AI and More
MongoDB | Blog
MongoDB | Blog

Hacker News: Show HN

PurrrrrFocus: Pomodoro Timer App - App Store Workflow Engine — Multi-Step Orchestration for Bun RapidPhoto: Pro Photo Editor App - App Store GitHub - DheerG/swarms: Achieve extraordinary results with claude code across a variety of tasks SPICE simulation → oscilloscope → verification with Claude Code — Lucas Gerads Show HN: VCoding – A 5 MB native Windows IDE with no dynamic dependencies Show HN: LLMs don't hallucinate because they're bad at math, it's the format GitHub - Agent-FM/agentfm-core: AgentFM is a peer-to-peer network that turns everyday computers into a decentralized AI supercomputer. AgentFM lets you run massive AI workloads directly across a global mesh of idle CPUs and GPUs. Show HN: Tracking Top US Science Olympiad Alumni over Last 25 Years GitHub - Potarix/agent-hub: One place to talk to all your agents Show HN: Runtime security for AI agents(injection,tool abuse, data exfiltration) GitHub - dubeyKartikay/lazyspotify: Terminal Spotify client for macOS and Linux GitHub - the-banana-tool/king-louie: Easy to use GUI Personal AI Assistant. Win/Linux/Mac. Show HN I made my vacation rental bookable by AI agents–no Airbnb, 0% commission GitHub - basteez/jsf-autoreload: maven plugin to enable hot reload on jsf projects uvm32/hosts/host-gdbstub at main · ringtailsoftware/uvm32 GitHub - labsai/EDDI: Config-driven engine that turns JSON into production-grade AI agents. Multi-agent orchestration, 12+ LLM providers, MCP/A2A protocols, RAG, persistent memory, and enterprise compliance (EU AI Act, GDPR, HIPAA). Built on Quarkus. GitHub - glitchnsec/fortyone-oss: AI Executive Assistant Platform Quickstart | Alien GitHub - muxshed/shed: One stream in, or many. Every destination, simultaneously. No cloud middleman, no per-channel fees, no limits. GitHub - ocrbase-hq/ocrbase: 📄 PDF/IMG ->.MD/JSON Document OCR API for PaddleOCR and GLMOCR. Self-hostable. GitHub - impactjo/home-memory: MCP server that lets your AI assistant remember everything about your home. GitHub - Sets88/dbcls: DbCls is a powerful terminal database client that supports various databases GitHub - neptun2000/heor-agent-mcp GitHub - SeanFDZ/macmind: Single-layer transformer in HyperTalk for the classic Macintosh RollQuation: Math Puzzles - Apps on Google Play GitHub - dropbox/witchcraft Show HN: Agent-cache – Multi-tier LLM/tool/session caching for Valkey and Redis GitHub - opentalon/opentalon: OpenTalon is an open-source platform built from the ground up in Go as a robust alternative to OpenClaw LinkedIn™ 职位抓取工具 - Chrome 应用商店
GitHub - unboundinnov/specdd: Spec-Driven Development as ...
diablovox · 2026-05-15 · via Hacker News: Show HN

Spec-Driven Development as a single Agent Skill. Drop it in, type naturally, ship features that don't embarrass you in two weeks.

specdd turns vague, top-of-mind feature requests into production-grade specs before code is written — without a CLI, without four config files, without ceremony you don't need. One skill, any modern AI coding agent: Claude (Code, Desktop, Web), Codex, Copilot, Cursor, Windsurf, Gemini CLI, anything else that reads SKILL.md or AGENTS.md-style instructions.

You: "build me a settings page where users can change their email"

specdd:
  → triages the scope (Feature)
  → asks the 2-3 questions you didn't think to address (verification email?
    active sessions? what's the success state?)
  → produces a tech-free spec + a 2-minute technical plan
  → only after you say "ship it" → writes code
  → runs the production-grade checklist before declaring done

No npx, no uv tool install, no specify init. Just a folder.


Why specdd exists

Two patterns broke the same way:

Vibe-coding — describe a feature in one line, get code that compiles and looks right but misses the empty state, the error state, the keyboard handling, the "what happens when the user double-clicks the button" — discovered in production.

Heavy spec frameworks like github/spec-kit — fix the rigor problem but introduce a CLI, four mandatory artifacts, and ceremony that's overkill for solo and one-off work.

specdd splits the difference. Spec-driven development that scales down to a 15-minute task and scales up to a multi-day project — same workflow, different ceremony.


What you get

  • One opinionated workflow — triage → interview → spec → plan → (tasks) → build & verify
  • Question-asking by default — interviews you for the gaps your one-liner left, instead of guessing
  • Tech-free specs — what & why for the user, separated from how for the agent
  • Production-grade checklist before "done" — empty/error/loading states, a11y, keyboard, mobile, observability
  • UX defaults baked in — five-states rule, confirmation patterns, optimistic updates, accessible focus management
  • Scales to scope — Quick (5-line inline mini-spec), Feature (spec + plan inline), Project (spec + plan + tasks as files)
  • Surface-agnostic — same skill in Claude Code, Claude.ai, Codex CLI, Cursor, Copilot, Windsurf

Install

Claude Code (recommended)

/plugin marketplace add unboundinnov/specdd
/plugin install specdd@specdd

That's it. The skill activates automatically when you ask for any feature work.

Claude Code (manual, no plugin system)

git clone --depth=1 https://github.com/unboundinnov/specdd /tmp/specdd-src
cp -r /tmp/specdd-src/plugins/specdd/skills/specdd ~/.claude/skills/
rm -rf /tmp/specdd-src

Or scope to a single project: copy into .claude/skills/ inside the project instead.

Claude.ai (Pro / Team / Enterprise / Max)

  1. Download specdd.zip from Releases (or zip plugins/specdd/skills/specdd/).
  2. Claude.ai → Settings → Capabilities → Skills → Upload skill.
  3. Upload the zip.

Codex CLI

mkdir -p ~/.codex/skills
git clone --depth=1 https://github.com/unboundinnov/specdd /tmp/specdd-src
cp -r /tmp/specdd-src/plugins/specdd/skills/specdd ~/.codex/skills/
rm -rf /tmp/specdd-src

Cursor / Copilot / Windsurf / Aider / others

These agents don't natively load SKILL.md yet. Copy the content of SKILL.md into the agent's instructions file:

  • Cursor: .cursor/rules/specdd.mdc (or legacy .cursorrules)
  • GitHub Copilot: .github/copilot-instructions.md
  • Windsurf: .windsurfrules
  • Aider: CONVENTIONS.md referenced from .aider.conf.yml

Reference files in references/ can be copied alongside or inlined.


How it works

Phase What happens
1. Triage Skip / Quick / Feature / Project — surfaced so you can correct it.
2. Interview 2–5 right questions for your request type, tappable where possible. Doesn't ask what it can infer.
3. Spec Tech-free: who / what / why / states / edge cases / out-of-scope / non-functional. You sign off.
4. Plan Tech approach: stack, files, decisions with rationale, alternatives, risks. <2 minutes to read.
5. Tasks (Project scope only) Ordered, verifiable units with explicit done-signals.
6. Build & verify Implement against spec; deviations require a spec update. Production checklist before "done".

SKILL.md is the entry point; templates, checklists, and examples are in references/.


specdd vs spec-kit

specdd github/spec-kit
Install Copy a folder, or one /plugin install uv tool install, specify init, CLI per-project
Surface Any agent that reads SKILL.md / instructions 30+ agents via Specify CLI
Ceremony Scales: 5 lines for a small change, 3 files for a project Mandatory: constitution.md, spec.md, plan.md, tasks.md
Slash commands None — conversational /speckit.specify, /speckit.plan, /speckit.tasks, /speckit.implement
UX defaults Baked in (5-states rule, a11y, confirmation patterns, motion) Project-defined in constitution.md
Best for Solo, indie, one-off features, mid-size teams Larger teams, established projects, formal SDD adoption

What specdd inherits from SDD literature (and spec-kit): the spec → plan → tasks workflow shape, the spec.md / plan.md / tasks.md filenames, the "tech-free spec" framing. These are general SDD concepts (Spec Driven Development pre-dates spec-kit), and keeping the filenames means specs migrate cleanly if you ever adopt spec-kit later.

What's specdd's own: the triage step (scale to scope), the interview phase with tappable options, the build & verify phase with the production-grade checklist, UX-first opinionation with the five-states rule, the common-sense gates. No CLI, no slash commands, no constitution.


Examples

See references/examples/:


Contributing

Issues and PRs welcome. Particularly interested in:

  • Request-type interview playbooks (the more specific, the better)
  • Surface-specific install instructions for new agents
  • Worked examples from real projects

Open an issue first for substantial changes; keep PRs focused; match the existing tone.


License

MIT. See LICENSE.


Credits

Inspired by, and respectfully different from, github/spec-kit, BMAD, and Kiro. Built on the Agent Skills format by Anthropic.

If specdd helps you ship better, a ⭐ helps people find it.