dont-reinvent - Teach your agent to acquire capabilities on its own - find existing skills before building one.

Mirrored in two places, same version everywhere: Hugging Face (you are here) · GitLab

Get the files - GitLab is the most reliable plain-git route:

git clone https://gitlab.com/LucioLiu/dont-reinvent.git

# or from this page
hf download LucioLiu/dont-reinvent --repo-type dataset --local-dir ./dont-reinvent

Then drop the folder into your agent's skills directory - ~/.claude/skills/dont-reinvent/ for Claude Code, ~/.agents/skills/dont-reinvent/ for Codex, or your client's equivalent.

Licence: PolyForm Noncommercial 1.0.0 - the LICENSE file in this repo is authoritative.

---

English | 简体中文

dont-reinvent · find the wheel before you build or redesign one 🎯

Teach an agent to reuse prior art before acquiring, creating, or substantially redesigning a capability.

What it solves

An agent without this discipline often does one of two things: say "I can't do that," or build from memory. A subtler failure happens when it already has a capability and rewrites that capability without checking what the ecosystem has learned since. Existing internal work is a valid execution wheel, but only a baseline when the wheel itself is being redesigned.

Install it, and the agent runs a six-step reuse routine for both capability gaps and substantial redesigns:

1. Classify the work — execution, routine no-behavior-change maintenance, or new/substantial capability design. Only the last category requires a broad prior-art gate by default.
2. Define and search the real problem — use outcomes, synonyms, adjacent categories, and anti-patterns, not only the current name. Inspect any user-named source directly and, in public/common domains, add at least two adjacent credible candidates.
3. Check the shelf, then look outside it when designing — "enough to execute" can stop at the shelf; "what should this wheel become?" cannot.
4. Compare every serious candidate on eight dimensions — source; license; permissions; complete tree; maintenance/obsolescence/update path; fit; money/data/time cost; human-present/unattended boundary.
5. Choose and record one reuse decision — adopt, adapt, compose, keep, or build; state what was and was not adopted and why.
6. Acquire or implement, register, and verify — substantial redesigns need preserved-behavior regression, changed-behavior forward tests, and a staleness check, not only structural validation.

Also covers three placement layers (user-level vs single-agent-level — attention isolation, so Role A's skills don't pollute Role B's session), freedom to adapt without bypassing prior-art review, offline/confidential honest degradation, and engineering safeguards such as complete-tree installation and full read-through before executing foreign instructions.

Honest positioning: this repo ships the reuse discipline itself — it does not ship a real provider, downloader, or installer, and the machine-readable provider / preview / provenance JSON Schemas are not published here yet (SKILL.md's eight-dimension comparison and its acquire/verify steps are the operative contract in the meantime). When the current harness lacks that runtime capability, it must name the missing coverage and produce a manual-review plan; documented steps must never be reported as an actual search, download, or install having happened.

Install

Option 1 · git clone (recommended, source-tracked and updatable)

git clone https://gitlab.com/LucioLiu/dont-reinvent ~/.claude/skills/dont-reinvent

Option 2 · Manual: copy SKILL.md to ~/.claude/skills/dont-reinvent/SKILL.md (no update channel — re-copy by hand when a new version ships). Chinese-speaking users should copy SKILL.zh-CN.md instead and rename it to SKILL.md locally.

Other tools: SKILL.md is a plain, portable markdown guide — feed it straight to Cursor / Codex / any LLM.

Why it should be your agent's first building block

It's the "entry point to every other capability" — install it once and your agent starts finding, evaluating, adapting, installing, and sedimenting capabilities on its own. More importantly, the same gate catches substantial rewrites of capabilities it already has.

Part of the capability-block architecture behind the Nuwa digital-employee framework — more blocks live in that repo (not in this folder).

Companion blocks (sibling skills, take what you need)

• token-saver — once you've found and installed a wheel, pick a model tier before dispatching a subagent; saves 70-90% of tokens on execution-type work.
• handoff-protocol — the full handoff discipline for multi-agent work delegation.

Version & updates

Current version 1.5.0 (see .claude-plugin/plugin.json). Updates are always published to this repo; git clone users run git pull in the installed directory to upgrade, manual-copy users should revisit this page and compare version numbers.

Verification boundary

The offline structural runner, the JSON Schemas it validates, and the SK-DR-P-001 / SK-DR-N-001 case bundles are not published in this repository yet — don't expect a tests directory in what you installed. Nothing here has been marked PASS for agent-semantic behavior: those cases stay not-run until an independent agent actually runs them and leaves a receipt.

License

PolyForm Noncommercial 1.0.0 — free for personal and non-commercial use.

---

Both language editions are maintained in sync. If they ever diverge, the Chinese edition is authoritative — please open an issue if you spot one.