[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-79962":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":15,"stars30d":16,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":17,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":18,"hasPages":18,"topics":20,"createdAt":9,"pushedAt":9,"updatedAt":21,"readmeContent":22,"aiSummary":23,"trendingCount":14,"starSnapshotCount":14,"syncStatus":15,"lastSyncTime":24,"discoverSource":25},79962,"helpline","coleam00\u002Fhelpline","coleam00","Demonstration codebase for building the AI Layer (CLAUDE.md hierarchy, hooks, skills, LSP, MCP, plugin) — companion repo for the 'How Claude Code works in large codebases' video.",null,"Python",89,29,77,0,2,10,4.43,false,"main",[],"2026-06-12 02:03:56","# Helpline\n\nHelpline is a B2B helpdesk and customer-support platform — an internal monorepo\nwhere five services (`api`, `auth`, `billing`, `notifications`, `search`) talk\nover HTTP and share two internal packages (`core`, `db`). It's the kind of\nmid-sized, multi-service codebase where an AI coding agent either pays for\nitself or quietly makes a mess: shared packages that everything imports, money\nrules that must never silently regress, a request gateway with its own\nconventions, and tests scattered across services.\n\n## Quick start\n\n**What this repo actually is.** Helpline is a *demonstration codebase*, built\nfor the YouTube video **\"The AI Layer: How to Make Claude Code Work in Large\nCodebases.\"** It is a deliberately realistic — but compact — complex codebase,\nused to show *concretely* how to build an **AI Layer** — our name for the\nharness of configuration and tooling that Anthropic's article describes in\n[*How Claude Code works in large codebases*](https:\u002F\u002Fclaude.com\u002Fblog\u002Fhow-claude-code-works-in-large-codebases-best-practices-and-where-to-start).\n(The article describes the harness and its components; \"AI Layer\" is our term\nfor it, not Anthropic's.)\nThat article is the *what* and the *why*, and it is excellent — but it stays\nhigh-level. Helpline is the *how*: every component from the article, built and\nvalidated on a real codebase. The rest of this repo is written as if Helpline\nwere a live product, because that is exactly how you should treat your own\ncodebase when you give it an AI Layer.\n\n**Run it:**\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fcoleam00\u002Fhelpline.git\ncd helpline\nuv sync --extra dev\nuv run pytest                 # the application's own test suite\n```\n\n**See the AI Layer:** read **[`AI-LAYER.md`](AI-LAYER.md)** — it maps every\ncomponent to the article and to where it lives in this repo. Then run the\nvalidator, which proves every piece works end to end:\n\n```bash\nuv run --extra dev python tooling\u002Fvalidate\u002Fvalidate_all.py   # 13\u002F13\n```\n\n## The AI Layer, concretely\n\nEvery codebase you ship with an AI agent has three parts: **code**, **tests**,\nand — the new one — the **AI Layer**: the configuration, context, and reusable\nworkflows that make an agent productive in *your* codebase. Helpline builds\nevery component the article names:\n\n| Component (from the article) | Built here | Where |\n|-------------------------------|------------|-------|\n| **CLAUDE.md hierarchy** | lean root `CLAUDE.md` + one per service\u002Fpackage, loaded additively | `CLAUDE.md`, `*\u002FCLAUDE.md` |\n| **Hooks** | a `SessionStart` orientation hook + a self-improving `Stop` hook that reflects on the session and proposes concrete `CLAUDE.md` edits | `.claude\u002Fhooks\u002F` |\n| **Skills** | `billing-money-rules`, `api-add-route`, `scoped-tests` — glob-scoped with the `paths:` field, progressive disclosure into `references\u002F` | `.claude\u002Fskills\u002F` |\n| **Subagent** | a genuinely read-only `explorer` (no write tools) that maps a subsystem and reports back | `.claude\u002Fagents\u002Fexplorer.md` |\n| **LSP** | pyright + `pyright-langserver` for symbol-level navigation | `docs\u002Flsp-setup.md` |\n| **MCP** | `codebase-search` — AST-based `where_is` \u002F `find_references` \u002F `outline` | `tooling\u002Fmcp\u002F` |\n| **Plugin** | `helpline-ai-layer` — bundles the portable pieces for one-command install | `tooling\u002Fhelpline-ai-layer\u002F` |\n\nThe git history is the before\u002Fafter, on purpose: **commit 1** is Helpline with\n*no* AI Layer; **commit 2** adds *the entire layer*.\n\n## Take this to your own codebase\n\nHelpline is not just something to read — it is something to *reuse*. There are\ntwo ways, and both matter more than the codebase itself.\n\n### 1. Install the plugin — the fastest path\n\nThe portable, repo-agnostic pieces of the AI Layer are bundled as a Claude Code\nplugin at `tooling\u002Fhelpline-ai-layer\u002F`. From your own project:\n\n```\n\u002Fplugin marketplace add \u002Fpath\u002Fto\u002Fhelpline\u002Ftooling\n\u002Fplugin install helpline-ai-layer@helpline-tooling\n```\n\nThat one install gives *your* repo:\n\n- the **self-improving Stop hook** — reflects on each session and proposes\n  `CLAUDE.md` updates, so your layer never silently rots;\n- the **read-only `explorer` subagent** — maps a subsystem without burning your\n  main session's context;\n- the **`codebase-search` MCP** — AST-based structured search;\n- a generic **`scoped-tests`** skill.\n\nThese are deliberately layout-agnostic: the hooks follow *your* `CLAUDE.md`\nhierarchy, and the MCP discovers *your* source by walking the repo. Nothing is\ntied to Helpline's directory names.\n\n### 2. Point your coding agent at this repo\n\nYou don't have to copy anything by hand. Clone Helpline, open it alongside your\nproject, and tell your agent something like:\n\n> Read `AI-LAYER.md` and the `.claude\u002F` folder in the Helpline repo. It is a\n> worked example of the AI Layer from Anthropic's large-codebases article.\n> Build a comparable AI Layer for *this* codebase — a CLAUDE.md hierarchy,\n> hooks, skills, an MCP, a subagent — adapted to our structure and conventions.\n\n`AI-LAYER.md` exists for exactly this. It is an agent-readable map from each\narticle concept to the artifact that implements it, plus the proof that it\nworks. The repo is a reference implementation your agent can learn the *pattern*\nfrom — not boilerplate to clone blindly.\n\n### What travels, and what you rebuild\n\n| Generalizable — works in any repo | Repo-specific — rebuild for yours |\n|-----------------------------------|------------------------------------|\n| the `Stop` \u002F `SessionStart` hooks (CLAUDE.md-hierarchy based) | the `CLAUDE.md` files (your conventions) |\n| the `explorer` subagent | `billing-money-rules`, `api-add-route` (your domain skills) |\n| the `codebase-search` MCP | — |\n| the generic `scoped-tests` skill | — |\n\nThe repo-specific pieces aren't reusable *content* — but they are reusable\n*templates*. Copy the shape, change the substance.\n\n## Layout\n\n```\nservices\u002F\n  api\u002F            HTTP gateway — routes, request handling\n  auth\u002F           authentication — tokens, password hashing\n  billing\u002F        subscriptions + invoicing\n  notifications\u002F  outbound email + templating\n  search\u002F         ticket search indexing + queries\npackages\u002F\n  core\u002F           shared domain models + error types\n  db\u002F             database connection + repositories\ninfra\u002F            docker-compose for local dev\nscripts\u002F          one-off operational scripts\ntests\u002F            cross-service integration tests\ntooling\u002F          the AI Layer's MCP server, the plugin, and the validators\n```\n\nEach service owns its own area. `packages\u002Fcore` and `packages\u002Fdb` are imported\nby every service — change them carefully.\n\n## Dev\n\n```bash\nuv sync --extra dev      # install\nuv run pytest            # full test suite\nuv run pyright           # type-check\n```\n","Helpline 是一个用于构建AI层的演示代码库，旨在展示如何在大型代码库中实现Claude AI编码代理。其核心功能包括CLAUDE.md层次结构、钩子、技能、LSP、MCP和插件等组件，并通过具体的代码实现Anthropic文章中的概念。项目采用Python语言编写，模拟了一个包含五个服务（api、auth、billing、notifications、search）的真实企业级帮助台平台。它适用于需要集成AI辅助开发工具到现有复杂多服务架构的企业场景，帮助开发者更好地理解和应用AI层技术于实际项目中。","2026-06-11 03:58:41","CREATED_QUERY"]