Research Before Build — AI skill for Claude Code
Research before build — a research-first meta-skill for AI agents: check prior art & community solutions before starting any non-trivial task.
How to install Research Before Build
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open pengkangzhen/research-before-build and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Research Before Build does
Research before build — a research-first meta-skill for AI agents: check prior art & community solutions before starting any non-trivial task.
Alternatives in AI
- Hindsight — State-of-the-art long-term memory for AI agents by Vectorize 6.7k ★
- Pixel Agents — Compressed Reference — VS Code extension with embedded React webview: pixel art office where AI agents (Claude Code terminals) are an 6.5k ★
- Develop — Development phase - Build solutions with multi-AI implementation and quality gates 2.8k ★
README
research-before-build
**Research before build — never reinvent the wheel.**
A meta-skill for AI coding agents that enforces a disciplined pre-flight check: before starting any non-trivial task, first search for **prior art and community solutions** — official docs, mature open-source libraries, GitHub issues, high-vote Q&A, papers, and internal knowledge bases.
先调研后构建——动手之前，先回答一个问题：**人类是否已经解决过这个问题？**
Why / 为什么需要
Reinventing the wheel is the single biggest waste in agent workflows. Community solutions are battle-tested, compressed experience. But research itself has a cost — so this skill gates research by task risk, trusts sources in tiers, and accepts research only when it changes a decision.
重新发明轮子是 agent 最大的浪费；社区方案是经过实战检验的压缩经验。 但调研本身也有成本，因此必须**分级触发、按信任采信、以决策影响为验收**。
Methodology / 方法论
一条公理、四层分级、五级信任、四维评估、一份契约、三道防线。
| Level | Task | Action |
|---|---|---|
| L0 | Trivial: typo, formatting, user-specified approach | Skip, execute directly |
| L1 | Routine coding | Local first: existing code, deps, docs, installed skills |
| L2 | High-risk / unfamiliar: deploy, migrate, obscure errors | Mandatory external research |
| L3 | Research tasks: surveys, architecture decisions | Research is the task |
Trust tiers: official docs → mature OSS → GitHub issues → high-vote Q&A → blogs/AI content (leads only, never evidence).
Every L2/L3 check must produce a **prior-art memo** with a decision (adopt / adapt / build), or explicitly state "no applicable prior art found (searched X/Y/Z)". Stop after two fruitless search rounds.
Install / 安装
npx skills add https://github.com/pengkangzhen/research-before-buildManual install (Codex):
git clone https://github.com/pengkangzhen/research-before-build
cp -r research-before-build/research-before-build ~/.agents/skills/Works with any agent that loads `SKILL.md`-style skills (Codex, Claude Code, Cursor, ...).
License
MIT
Related Skills
Askproof Skill
Verify AI agent output with actionable evidence for non-engineers. Build better prompts by auditing AI claims
Prior Check
Generate and analyze prior predictive checks for the current model
Aurora
Aurora - independent community smart home orchestrator. Routes any smart home request to the right specialist
Swarm Plan
Route ANY subagent-spawning task to the right vector, tier, and pins before spawning anything — the general pr
Update Model Routing
Deep-research current public model benchmarks and refresh MODEL-ROUTING.md (the advisory per-task routing tabl
Animated Drawing
Animate a SUPPLIED drawing/photo of a character with real motion capture (Meta AnimatedDrawings) → raster GIF/
Related Agents
History Reviewer
Git history-aware reviewer. Use when changes touch code with non-trivial history, or to check whether a fix ac
Hydra Orchestrator
Route a non-trivial goal to the right Hydra knowledge, skills, workflows, and task state. Use when work spans
Worktree Setup.Agent
Use when starting work on any non-trivial task that involves edits, multiple parallel tasks on the same repo,
