← ClaudeAtlas

oss-first-developmentlisted

Before writing a complex system from scratch, find, evaluate, reuse and adapt existing open-source code, libraries, tools and MCP servers, then build only the glue or the improvement. Uses targeted web dorks (site:, exact phrases, file formats, API symbols) chained on identifiers from each result, optionally a long-tail GitHub topic sweep and user-supplied seed repos; verifies repos contain source, checks licenses (MIT/Apache vs GPL/AGPL/none/non-commercial), labels each project base / learn-from / noise, and writes a reuse-adapt-idea-or-write decision, clone script and build plan. Use when the user wants to build a complex app, tool or backend, asks "is there a library/repo for X", "kya koi project hai", wants to avoid reinventing the wheel, compare existing implementations, convert A to B (SVG to Compose), clone an existing product, or says "research first" or "don't write from scratch".
zohaibmushwani/agent-skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill zohaibmushwani/agent-skills
# oss-first-development Principle: **look at what exists before you write anything; write only what does not exist or what you can do better.** A complex system assembled from proven parts is faster, less buggy and easier to defend than a rewrite - and "better than existing" becomes a concrete, checkable claim. ## The one method: dork, then chain Five real research runs (see `examples/`) tested three discovery methods on the same topic and judged every result. Web dorks won: **~95% of what they returned was relevant** and they surfaced the projects that were actually usable as bases. Repo search by GitHub topic (~19%) and public code search (~15%) returned mostly noise or repos with no source. So this skill uses **one** discovery method and keeps the others only as helpers: - **Discovery = targeted web dorks + identifier chaining** (this skill's core, below). - Code search (`scripts/code_chain.py`, `gh search code`) is **not** for discovery; use it only *inside an already shortlisted repo* to find the exact file. - Regional-language queries are added **only** when the ecosystem itself is regional (MMD -> JA/CN, Shadowsocks/Xray -> CN/RU); otherwise they just repeat English hits. ## Workflow Work in `research/<topic-slug>/` (`scripts/scaffold_kit.py <topic>` creates it). Keep notes in files; later phases read notes instead of re-reading repos. 1. **Map.** Split the goal into 3-6 *pillars*. For each write: plain words, synonyms, **API symbols** (`getFft`, `glUniform1f`), *