codex-arsenal 0.2.0 → 0.3.0

package/CODEX.md

@@ -91,7 +91,20 @@ Weak criteria ("make it work") require constant clarification.
 
 ---
 
-## 6. Context Awareness
+## 6. Token Budgeting
+
+**Spend context only where it changes the decision.**
+
+- Start with filenames, symbols, and short snippets before reading full files.
+- Prefer `rg` and targeted search over broad recursive reads.
+- Summarize long outputs immediately: key facts, file paths, line numbers, open questions.
+- Do not paste large logs back to the user. Quote only the failing line or smallest useful excerpt.
+- Reuse already-loaded context instead of re-reading the same files.
+- When handing off, compress to: goal, changed files, commands run, verification status, next risk.
+
+---
+
+## 7. Context Awareness
 
 **Codex has a limited context window — use it wisely.**
 
@@ -102,7 +115,7 @@ Weak criteria ("make it work") require constant clarification.
 
 ---
 
-## 7. Tool Use Discipline
+## 8. Tool Use Discipline
 
 **Every tool call has a cost. Make them count.**
 
@@ -118,4 +131,4 @@ Weak criteria ("make it work") require constant clarification.
 - Diffs contain fewer unnecessary changes
 - Fewer rewrites due to overcomplication
 - Clarifying questions come **before** implementation, not after mistakes
-- Agentic runs complete tasks end-to-end without silent failures or scope creep
+- Agentic runs complete tasks end-to-end without silent failures or scope creep

package/README.md

@@ -1,173 +1,189 @@
 # Codex-Arsenal
 
-> OpenAI Codex / Claude Code 기반 개발 경험을 극대화하기 위한  
-> 플러그인 · 스킬 · 워크플로 · 설정 · 프롬프트 큐레이션 레포지토리.
+Practical building blocks for working with OpenAI Codex.
 
----
+Codex-Arsenal is not an awesome list. It is a small, installable collection of Codex guidelines, skills, prompts, workflows, configs, and plugin sketches that can be copied into real projects.
 
-## 🧭 이 레포는 무엇인가?
+## Install
 
-**Codex-Arsenal**은 AI 코딩 에이전트(OpenAI Codex, Claude Code 등)를 실무에 적용하면서  
-"vanilla setup만으로는 부족하다"고 느낀 개발자들을 위한 실용 자료 모음입니다.
+Run the CLI without installing it globally:
 
-단순한 awesome list와 다른 점:
-
-- 실제로 써봤거나, 체감 생산성이 검증된 것만 수록
-- setup → 사용법 → 주의사항까지 한 곳에서 해결
-- 개인 개발자 / 소규모 팀 모두를 대상으로 함
-- Claude Code 스타일의 agentic workflow에 최적화
+```bash
+npx codex-arsenal list
+```
 
-> **대상 독자:** 이미 Codex 또는 Claude Code를 써본 개발자.  
-> AI 코딩 도구가 처음이라면 [공식 문서](https://platform.openai.com/docs/guides/code)부터 시작하세요.
+Install the default item set into the current project:
 
----
+```bash
+npx codex-arsenal init --yes
+```
 
-## 📂 레포지토리 구조
+Install specific items:
 
+```bash
+npx codex-arsenal get codex-md skill-publishing-npm-packages
 ```
-codex-arsenal/
-├── plugins/          # 에이전트 기능을 확장하는 플러그인 모음
-├── skills/           # 재사용 가능한 에이전트 스킬 패키지
-├── prompts/          # 프롬프트 템플릿 및 system prompt 예제
-├── workflows/        # 워크플로 정의 및 실행 스크립트
-├── configs/          # .codex, VSCode, zsh, tmux 등 설정 파일
-├── references/       # 논문 · 블로그 · 주목할 GitHub 레포 링크
-├── examples/         # 실전 데모 및 before/after 예제
-└── README.md
+
+Install into another directory:
+
+```bash
+npx codex-arsenal get codex-md --dir ./my-project
 ```
 
----
+On Windows or when running from a package directory with the same name, this form is the most reliable:
 
-## ✨ 수록 콘텐츠 카테고리
+```bash
+npm exec --yes --package=codex-arsenal -- codex-arsenal list
+```
 
-### 🔌 Plugins
+## CLI
 
-에이전트 자체 기능을 보강하는 플러그인.
+```bash
+codex-arsenal init [--yes] [--force] [--dir <path>]
+codex-arsenal list
+codex-arsenal get <id...> [--force] [--dir <path>]
+```
 
-현재 수록 중 / 예정:
+- `init` opens a small selector. With `--yes`, it installs default items without prompting.
+- `list` prints all installable manifest entries grouped by category.
+- `get` installs one or more manifest entries by id.
+- Existing files are skipped by default. Add `--force` to overwrite them intentionally.
 
-| 이름 | 설명 | 상태 |
-|------|------|------|
-| context-window-compressor | 긴 대화 컨텍스트를 압축해 토큰 절약 | 준비 중 |
-| git-diff-helper | PR diff 요약 및 리뷰 요청 자동화 | 준비 중 |
-| multi-agent-router | 여러 에이전트에 작업을 분배하는 라우터 | 계획 중 |
-| repo-memory | 프로젝트 구조를 기억하는 메모리 레이어 | 계획 중 |
+The package is published on npm as [`codex-arsenal`](https://www.npmjs.com/package/codex-arsenal).
 
----
+## What Is Included
 
-### 🧠 Skills
+### Behavior Guidelines
 
-반복 작업에서 즉시 꺼내 쓸 수 있는 스킬셋.
+| ID | Installs | Purpose |
+| --- | --- | --- |
+| `codex-md` | `CODEX.md` | Project-local behavioral guardrails for Codex-style agents. |
 
-| 스킬 | 설명 |
-|------|------|
-| `debug-workflow` | 에러 로그 → 원인 분석 → 수정 → 재현 테스트까지 파이프라인화 |
-| `pr-reviewer` | PR diff를 받아 리뷰 코멘트를 자동 생성 |
-| `test-gen` | 함수 시그니처와 코드를 보고 유닛 테스트 자동 생성 |
-| `refactor-pipeline` | 코드 품질 분석 → 리팩터링 제안 → 적용까지 단계 처리 |
-| `doc-gen` | 코드베이스에서 README, API 문서 자동 작성 |
-| `arch-analysis` | 레포 전체 구조를 분석해 의존성 맵 생성 |
+### Configs
 
----
+| ID | Installs | Purpose |
+| --- | --- | --- |
+| `config-codex` | `.codex/config.json` | Starter Codex config. |
+| `config-vscode` | `.vscode/settings.json` | VS Code settings for agent-assisted development. |
 
-### ⚙️ Configs
+### Prompts
 
-실제 사용 중인 설정 파일 모음. 복붙 후 바로 사용 가능한 형태로 제공.
+| ID | Installs | Purpose |
+| --- | --- | --- |
+| `prompt-solo-dev` | `prompts/system-prompts/solo-dev.md` | Solo developer system prompt template. |
 
-- `.codex/config.json` — Codex 에이전트 기본 설정 프리셋
-- `settings.json` — VSCode + Codex 통합 설정
-- `system-prompts/` — 역할별 system prompt 템플릿 (solo dev / 팀 / 레거시 분석)
-- `.zshrc` 스니펫 — Codex CLI alias 및 단축키 모음
-- `tmux.conf` — AI 세션을 병렬로 띄우기 위한 tmux 레이아웃
+### Skills
 
----
+| ID | Installs | Purpose |
+| --- | --- | --- |
+| `skill-debug-workflow` | `skills/debug-workflow/` | Reproduce, diagnose, test, and fix bugs systematically. |
+| `skill-test-gen` | `skills/test-gen/` | Generate focused tests from behavior notes and function signatures. |
+| `skill-publishing-npm-packages` | `skills/publishing-npm-packages/SKILL.md` | Prepare, troubleshoot, and automate npm releases with Trusted Publishing. |
 
-### 🧩 Workflows
+### Plugins
 
-"어떻게 써야 잘 쓰는가"에 대한 실전 운영 방식.
+| ID | Installs | Purpose |
+| --- | --- | --- |
+| `plugin-context-window-compressor` | `plugins/context-window-compressor/` | Plugin sketch for compressing long agent context into concise handoff notes. |
 
-| 워크플로 | 대상 | 핵심 아이디어 |
-|----------|------|---------------|
-| `solo-dev-loop` | 혼자 개발하는 개발자 | 계획 → 코딩 → 테스트 → 커밋을 에이전트와 함께 |
-| `startup-mvp` | 빠른 프로토타입이 필요한 팀 | 스펙 → 구현 → 배포 최단 경로 |
-| `legacy-onboarding` | 오래된 코드베이스 합류 | 에이전트로 구조 파악 → 코드 맵 생성 |
-| `ai-pair-programming` | 주니어/미드 개발자 | 에이전트를 시니어처럼 활용하는 패턴 |
-| `multi-agent-orch` | 복잡한 대형 태스크 | 여러 에이전트에게 역할 분배 후 결과 통합 |
+### Workflows
 
----
+| ID | Installs | Purpose |
+| --- | --- | --- |
+| `workflow-solo-dev-loop` | `workflows/solo-dev-loop/` | A plan, code, test, commit loop for solo developers using agents. |
 
-### 📚 References
+## Repository Layout
 
-읽어볼 가치 있는 자료 큐레이션. 링크만 나열하지 않고, **왜 읽어야 하는지** 한 줄 설명 포함.
+```text
+codex-arsenal/
+  bin/                  CLI entrypoint
+  lib/                  manifest, installer, and fetcher
+  configs/              reusable editor and agent configs
+  plugins/              plugin sketches
+  prompts/              system prompt templates
+  skills/               reusable agent skills
+  workflows/            repeatable agentic workflows
+  references/           curated references and notes
+  examples/             before/after examples
+  test/                 Node test suite
+```
 
-- 논문 · 연구 자료
-- 주목할 GitHub 레포 (with star 기준 및 실용성 코멘트)
-- 블로그 · 튜토리얼 (검증된 것만)
-- 프롬프트 엔지니어링 컬렉션
+`lib/manifest.js` is the source of truth for installable items. Add new content there when you want it to appear in `codex-arsenal list` or be installable through `codex-arsenal get`.
 
----
+## Development
 
-## ⭐ 수록 기준
+Run tests:
 
-모든 콘텐츠는 아래 기준을 만족해야 합니다:
+```bash
+npm test
+```
 
-| 기준 | 설명 |
-|------|------|
-| ✅ 실사용 검증 | 실제로 사용해봤거나, 커뮤니티에서 반복 검증된 것 |
-| ✅ 실무 적용 가능 | 장난감 예제가 아닌 실제 코드베이스에 적용 가능 |
-| ✅ 반복 사용성 | 한 번만 쓰고 마는 것이 아닌, 루틴으로 쓸 수 있는 것 |
-| ✅ setup 대비 효율 | 설치/설정 비용보다 얻는 생산성이 명확히 큰 것 |
-| ✅ 한계 보완 | Codex/Claude Code의 알려진 약점을 보완하는 것 |
+Preview the npm package contents:
 
----
+```bash
+npm pack --dry-run
+```
 
-## 🛠 빠른 시작
+Try the local CLI:
 
 ```bash
-# 레포 클론
-git clone https://github.com/your-username/codex-arsenal.git
-cd codex-arsenal
+node bin/cli.js list
+node bin/cli.js get codex-md --dir ./tmp-install
+```
 
-# 설정 파일 적용 (예시: VSCode 설정)
-cp configs/vscode/settings.json ~/.vscode/settings.json
+## Publishing
 
-# 워크플로 스크립트 실행 (예시: solo-dev-loop)
-./workflows/solo-dev-loop/setup.sh
+This repository is configured for npm Trusted Publishing through GitHub Actions.
+
+Release flow:
+
+```bash
+npm test
+npm pack --dry-run
+npm version patch
+git push --follow-tags
 ```
 
-> 각 폴더에는 독립적인 `README.md`가 있어 개별 설치 및 사용법을 안내합니다.
+Use `minor` instead of `patch` when adding new installable content or CLI behavior:
+
+```bash
+npm version minor
+git push --follow-tags
+```
 
----
+The publish workflow runs on `v*` tags and publishes with OIDC, so it does not require a long-lived `NPM_TOKEN`.
 
-## 🤝 기여 방법
+## Codex Plugin Distribution
 
-좋은 자료가 있다면 PR로 기여해주세요. 특히 환영하는 기여:
+The npm package is the current public distribution channel. Codex plugin library publishing is separate from npm and currently does not expose a general public submission flow in the public docs.
 
-- 실제로 오래 써서 검증된 것
-- before / after 생산성 차이가 명확한 것
-- setup 과정이 step-by-step으로 정리된 것
-- 특정 상황(레거시, 대형 레포, 팀 협업 등)에서 효과적인 것
+For now, use one of these paths:
 
-**기여 절차:**
+- publish reusable files through `codex-arsenal` on npm;
+- keep plugin sketches under `plugins/`;
+- create a local or team marketplace file at `.agents/plugins/marketplace.json` when you want Codex app users to install a repo plugin from a local marketplace;
+- track OpenAI's Codex plugin documentation for an official public marketplace submission path.
 
-1. `fork` → feature 브랜치 생성 (`feat/plugin-이름` 또는 `feat/workflow-이름`)
-2. 해당 카테고리 폴더에 파일 추가 (각 폴더의 `TEMPLATE.md` 참고)
-3. PR 제목에 카테고리 명시: `[plugin] context-window-compressor 추가`
-4. PR 본문에 **사용 배경 / 실제 효과 / 주의사항** 포함
+See `docs/codex-plugin-distribution.md` for the current assessment.
 
----
+## Contribution Guidelines
 
-## 🌌 방향성
+Good additions should be:
 
-가까운 미래의 개발 환경은 "IDE + 자동완성 AI" 수준을 넘어,  
-개발자가 **여러 AI 에이전트를 orchestration하며 일하는 구조**에 가까워질 것입니다.
+- useful in real projects, not just demos;
+- small enough to understand quickly;
+- documented with setup and usage notes;
+- installable through the manifest when appropriate;
+- verified with tests or a concrete manual check.
 
-이 레포는 그 방향으로의 전환을 실험하고,  
-실제로 쓸 수 있는 도구와 패턴을 기록하는 공간입니다.
+When adding a new installable item:
 
----
+1. Add the files under the appropriate top-level directory.
+2. Add a manifest entry in `lib/manifest.js`.
+3. Run `npm test`.
+4. Run `node bin/cli.js list` and confirm the item appears.
+5. Run `npm pack --dry-run` and confirm the intended files are included.
 
-## 📌 태그
+## License
 
-`codex` `claude-code` `ai-agent` `developer-tools` `prompt-engineering`  
-`llm-workflow` `productivity` `automation` `vscode` `terminal`
+MIT

package/bin/cli.js

@@ -6,9 +6,11 @@ function printHelp() {
   console.log(`codex-arsenal
 
 Usage:
-  codex-arsenal init [--yes] [--dir <path>]
+  codex-arsenal init [--yes] [--force] [--dir <path>]
   codex-arsenal list
-  codex-arsenal get <id...> [--dir <path>]
+  codex-arsenal get <id...> [--force] [--dir <path>]
+
+By default, existing files are skipped. Use --force to overwrite them.
 `);
 }
 
@@ -28,7 +30,7 @@ function nonOptionArgs(args) {
       index += 1;
       continue;
     }
-    if (arg === "--yes" || arg === "-y") {
+    if (arg === "--yes" || arg === "-y" || arg === "--force") {
       continue;
     }
     result.push(arg);
@@ -51,6 +53,7 @@ function printList() {
 async function main(argv) {
   const [command = "init", ...args] = argv;
   const dir = readOption(args, "--dir", readOption(args, "-d", process.cwd()));
+  const force = args.includes("--force");
 
   if (command === "help" || command === "--help" || command === "-h") {
     printHelp();
@@ -67,12 +70,12 @@ async function main(argv) {
     if (!ids.length) {
       throw new Error("get requires at least one item id");
     }
-    await runInit({ preSelected: findManifestItems(ids), skipPrompt: true, dir });
+    await runInit({ preSelected: findManifestItems(ids), skipPrompt: true, dir, force });
     return;
   }
 
   if (command === "init") {
-    await runInit({ yes: args.includes("--yes") || args.includes("-y"), dir });
+    await runInit({ yes: args.includes("--yes") || args.includes("-y"), dir, force });
     return;
   }
 

package/docs/codex-plugin-distribution.md

@@ -0,0 +1,54 @@
+# Codex Plugin Distribution
+
+## Current Assessment
+
+As of May 23, 2026, the practical public distribution path for Codex-Arsenal is npm. OpenAI's public Codex material says Codex users can browse the plugins library and create plugins in the Codex app, but it does not document a general public submission process for third-party plugins.
+
+Relevant OpenAI pages:
+
+- [Plugins and skills](https://openai.com/academy/codex-plugins-and-skills/)
+- [Codex](https://openai.com/codex/)
+- [Using Codex with your ChatGPT plan](https://help.openai.com/en/articles/11369540/)
+
+## Recommended Path
+
+1. Keep publishing installable skills, prompts, configs, and workflow files through `codex-arsenal` on npm.
+2. Keep plugin ideas in `plugins/<name>/` until they become real Codex plugins.
+3. When a plugin is ready for Codex app testing, convert it to a Codex plugin structure with `.codex-plugin/plugin.json`.
+4. Add a repo-local marketplace entry in `.agents/plugins/marketplace.json` for team/local installation.
+5. If OpenAI opens a formal public marketplace submission path, package the plugin manifest, documentation, permission model, and test notes for review.
+
+## Repo Marketplace Shape
+
+Use this shape for local or team discovery:
+
+```json
+{
+  "name": "codex-arsenal",
+  "interface": {
+    "displayName": "Codex-Arsenal"
+  },
+  "plugins": [
+    {
+      "name": "context-window-compressor",
+      "source": {
+        "source": "local",
+        "path": "./plugins/context-window-compressor"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity"
+    }
+  ]
+}
+```
+
+## Before Submitting Anywhere
+
+- Define exactly what external systems the plugin touches.
+- Keep permissions narrow and explain authentication.
+- Provide a small demo workflow and screenshots if the plugin has UI.
+- Include test notes for install, auth, failure states, and uninstall.
+- Avoid bundling broad arsenals into one plugin. Publish focused plugins that solve one job well.

package/docs/npm-publishing.md

@@ -0,0 +1,35 @@
+# npm Publishing
+
+This package is set up for npm Trusted Publishing from GitHub Actions.
+
+## One-time npm setup
+
+1. Create or claim the `codex-arsenal` package on npm.
+2. On npmjs.com, open the package settings and add a Trusted Publisher:
+   - Provider: GitHub Actions
+   - Owner: `citron03`
+   - Repository: `Codex-Arsenal`
+   - Workflow filename: `publish.yml`
+   - Environment: `npm`
+3. After the first trusted publish works, set publishing access to require 2FA and disallow tokens.
+
+## Release process
+
+1. Ensure `main` is green.
+2. Update the version:
+
+   ```bash
+   npm version patch
+   ```
+
+3. Push the commit and tag:
+
+   ```bash
+   git push --follow-tags
+   ```
+
+GitHub Actions will run tests, perform a package dry run, and publish to npm from the `v*` tag.
+
+## Why Trusted Publishing
+
+Trusted Publishing uses short-lived OIDC credentials from GitHub Actions instead of a long-lived `NPM_TOKEN`. For public packages published from public repositories, npm also creates provenance attestations automatically.

package/lib/installer.js

@@ -2,12 +2,15 @@ import { mkdir, readFile, writeFile } from "node:fs/promises";
 import { createInterface } from "node:readline/promises";
 import { stdin as input, stdout as output } from "node:process";
 import path from "node:path";
+import { fileURLToPath } from "node:url";
 import { BASE_URL, MANIFEST } from "./manifest.js";
 import { fetchFile } from "./fetcher.js";
 
+const PACKAGE_ROOT = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
+
 async function readSourceFile(src) {
   try {
-    return await readFile(src, "utf8");
+    return await readFile(path.join(PACKAGE_ROOT, src), "utf8");
   } catch (error) {
     if (error.code !== "ENOENT") {
       throw error;
@@ -16,28 +19,45 @@ async function readSourceFile(src) {
   }
 }
 
-export async function installItems(items, targetDir) {
+export async function installItems(items, targetDir, opts = {}) {
   const resolvedTarget = path.resolve(targetDir);
   let installed = 0;
   let failed = 0;
+  const successes = [];
+  const skipped = [];
   const failures = [];
+  const readSource = opts.readSourceFile || readSourceFile;
+  const force = !!opts.force;
 
   for (const item of items) {
     for (const file of item.files) {
       const destination = path.join(resolvedTarget, file.dest);
       try {
-        const content = await readSourceFile(file.src);
+        if (!force) {
+          try {
+            await readFile(destination, "utf8");
+            skipped.push(file.dest);
+            continue;
+          } catch (error) {
+            if (error.code !== "ENOENT") {
+              throw error;
+            }
+          }
+        }
+
+        const content = await readSource(file.src);
         await mkdir(path.dirname(destination), { recursive: true });
         await writeFile(destination, content, "utf8");
         installed += 1;
+        successes.push(file.dest);
       } catch (error) {
         failed += 1;
-        failures.push({ item: item.id, file: file.src, error: error.message });
+        failures.push({ item: item.id, file: file.src, dest: file.dest, error: error.message });
       }
     }
   }
 
-  return { installed, failed, failures, targetDir: resolvedTarget };
+  return { installed, failed, successes, skipped, failures, targetDir: resolvedTarget };
 }
 
 function defaultItems() {
@@ -75,15 +95,24 @@ export async function runInit(opts = {}) {
 
   if (!selected.length) {
     console.log("No items selected.");
-    return { installed: 0, failed: 0, failures: [], targetDir: path.resolve(opts.dir || ".") };
+    return { installed: 0, failed: 0, successes: [], skipped: [], failures: [], targetDir: path.resolve(opts.dir || ".") };
   }
 
-  const result = await installItems(selected, opts.dir || process.cwd());
-  for (const item of selected) {
-    for (const file of item.files) {
-      console.log(`installed ${file.dest}`);
-    }
+  const result = await installItems(selected, opts.dir || process.cwd(), {
+    readSourceFile: opts.readSourceFile,
+    force: opts.force
+  });
+  for (const file of result.successes) {
+    console.log(`installed ${file}`);
+  }
+  for (const file of result.skipped) {
+    console.log(`skipped ${file} (already exists; use --force to overwrite)`);
+  }
+  for (const failure of result.failures) {
+    console.log(`failed ${failure.dest} (${failure.error})`);
   }
-  console.log(`\nDone: ${result.installed} installed${result.failed ? `, ${result.failed} failed` : ""}.`);
+  console.log(
+    `\nDone: ${result.installed} installed${result.skipped.length ? `, ${result.skipped.length} skipped` : ""}${result.failed ? `, ${result.failed} failed` : ""}.`
+  );
   return result;
 }

package/lib/manifest.js

@@ -1,4 +1,4 @@
-export const REPO = "your-username/codex-arsenal";
+export const REPO = "citron03/Codex-Arsenal";
 export const BRANCH = "main";
 export const BASE_URL = `https://raw.githubusercontent.com/${REPO}/${BRANCH}`;
 
@@ -11,13 +11,6 @@ export const MANIFEST = [
     files: [{ src: "CODEX.md", dest: "CODEX.md" }],
     default: true
   },
-  {
-    id: "claude-md",
-    category: "Behavior Guidelines",
-    label: "CLAUDE.md",
-    description: "Behavior guidelines adapted for Claude Code agents.",
-    files: [{ src: "CLAUDE.md", dest: "CLAUDE.md" }]
-  },
   {
     id: "config-codex",
     category: "Configs",

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "codex-arsenal",
-  "version": "0.2.0",
-  "description": "A practical arsenal of Codex and Claude Code plugins, skills, prompts, workflows, and configs.",
+  "version": "0.3.0",
+  "description": "A practical arsenal of Codex plugins, skills, prompts, workflows, and configs.",
   "type": "module",
   "bin": {
     "codex-arsenal": "bin/cli.js"
@@ -22,9 +22,10 @@
   },
   "files": [
     "bin/",
+    "docs/codex-plugin-distribution.md",
+    "docs/npm-publishing.md",
     "lib/",
     "CODEX.md",
-    "CLAUDE.md",
     "configs/",
     "plugins/",
     "prompts/",
@@ -36,7 +37,6 @@
   },
   "keywords": [
     "codex",
-    "claude-code",
     "ai-agent",
     "developer-tools",
     "prompt-engineering"

package/CLAUDE.md

@@ -1,15 +0,0 @@
-# CLAUDE.md
-
-Use this file as a project-local behavior guide for Claude Code agents.
-
-## Operating Principles
-
-- Read the repository instructions before editing.
-- Keep changes small and directly tied to the task.
-- Prefer tests before behavior changes.
-- Explain uncertainty instead of guessing silently.
-- Do not overwrite user work unless explicitly asked.
-
-## Completion Criteria
-
-Before reporting completion, run the smallest command that proves the change works and summarize the result.