instrlint 0.2.2 → 0.2.4

package/README.md

@@ -8,24 +8,33 @@
 
 **instrlint lints itself.**
 
-Lint and optimize your `CLAUDE.md` / `AGENTS.md` — find dead rules, token waste, and structural issues. One command, one score, one report.
+## Why instrlint?
 
-## Quick Start
+Your `CLAUDE.md` grows over time. Rules get copied from templates, duplicated across sections, or left pointing at files that no longer exist. Some rules repeat what `tsconfig.json` or `.editorconfig` already enforces — burning token budget with no benefit.
 
-**Global install (recommended):**
+instrlint scans your instruction files and tells you exactly what to fix. One command, one score, one actionable report.
 
-```bash
-npm install -g instrlint
-instrlint
-```
+Two things make it different from a line counter:
 
-Install once, run from any project. Or use `npx` to run without installing:
+- **Host-orchestrated LLM verification** — run `/instrlint --verify` in Claude Code and the host model judges ambiguous findings itself. No API key. No separate service. The model already running your session *is* the verifier.
+- **Refactoring walkthrough** — when your root file is too long, instrlint doesn't just say "split this." It walks you through each section, classifies it into one of four buckets, and guides the decision — because not every long CLAUDE.md should be split the same way.
+
+Supports Claude Code, Codex, and Cursor. 487 tests. Dogfoods itself since v0.2.3.
+
+## Quick start
 
 ```bash
 npx instrlint
 ```
 
-Run from your project root where `CLAUDE.md` or `AGENTS.md` lives.
+Run from the directory where your `CLAUDE.md` or `AGENTS.md` lives. No install required.
+
+Or install globally and run from any project:
+
+```bash
+npm install -g instrlint
+instrlint
+```
 
 ## Output example
 
@@ -59,19 +68,17 @@ Run from your project root where `CLAUDE.md` or `AGENTS.md` lives.
   ── 1 critical · 9 warnings · 12 suggestions ────────
 ```
 
-## Features
+## What it finds
 
-**Token Budget** — counts tokens across all your instruction files using `cl100k_base` encoding (falls back to character estimation). Flags files over 200 lines and baseline context over 25% of your window.
+**Token budget** — measures tokens across all instruction files and MCP servers using `cl100k_base` encoding. Flags root files over 200 lines and baselines above 25% of your context window.
 
-**Dead Rules** — detects rules already enforced by your config files. ~15 overlap patterns covering TypeScript strict mode, Prettier formatting, ESLint rules, conventional commits, EditorConfig, and more.
+**Dead rules** — rules your config files already enforce: TypeScript strict mode, Prettier formatting, ESLint, EditorConfig, conventional commits, C# `.csproj`/`.editorconfig` settings, and more. ~20 patterns covering JS/TS and C# ecosystems.
 
-**Structure** — finds contradicting rules, stale file references, exact and near-duplicate rules, and path-scoping opportunities.
+**Structural issues** — contradicting rules, stale file references, near-duplicate rules, and path-scoping opportunities.
 
-**Auto-fix** — `--fix` safely removes dead rules, stale refs, and exact duplicates. Requires a clean git working tree.
+**Auto-fix** — `--fix` removes dead rules, stale refs, and exact duplicates. Requires a clean git working tree.
 
-**LLM Verification** — `/instrlint --verify` (Claude Code / Codex skill) runs a two-pass protocol: instrlint emits low-confidence findings as `candidates.json`, you (the host agent) judge each one and write `verdicts.json`, then instrlint merges results — filtering false positives and attaching `✓ confirmed` / `❓ uncertain` badges. No API key required; the host model is the verifier.
-
-**CI Integration** — `instrlint ci` exits 0 or 1 based on findings severity, with SARIF output for GitHub Code Scanning.
+**CI integration** — `instrlint ci` exits non-zero based on severity. SARIF output for GitHub Code Scanning. Generate a workflow with `instrlint init-ci --github`.
 
 ## Supported tools
 
@@ -83,112 +90,58 @@ Run from your project root where `CLAUDE.md` or `AGENTS.md` lives.
 
 Auto-detected from project structure. Override with `--tool`.
 
-## Commands
-
-```bash
-instrlint                         # Full health check — score, grade, top issues
-instrlint budget                  # Token budget analysis only
-instrlint deadrules               # Dead rule detection only
-instrlint structure               # Structural analysis only
-instrlint --fix                   # Auto-fix safe issues
-instrlint --fix --force           # Skip git clean check
-
-instrlint --format json           # JSON output (for CI, scripts)
-instrlint --format markdown       # Markdown output (for PR comments)
-instrlint --lang zh-TW            # Traditional Chinese output
-
-instrlint ci                      # CI mode: exit 1 on critical findings
-instrlint ci --fail-on warning    # Exit 1 on warnings too
-instrlint ci --format sarif       # SARIF output for GitHub Code Scanning
-instrlint ci --output report.sarif  # Write to file
-
-instrlint init-ci --github        # Generate .github/workflows/instrlint.yml
-instrlint init-ci --gitlab        # Print GitLab CI snippet to stdout
-
-instrlint install --claude-code   # Install as global Claude Code skill
-instrlint install --claude-code --project  # Install into project
-instrlint install --codex         # Install as Codex skill
-
-# Host-orchestrated LLM verification (two-pass, no API key needed)
-instrlint --emit-candidates instrlint-candidates.json        # Write low-confidence findings for host LLM to judge
-instrlint --emit-candidates instrlint-candidates.json --skip-report  # Same, without printing the report
-instrlint --apply-verdicts instrlint-verdicts.json           # Apply host LLM verdicts and re-render report
-```
-
-## CI Integration
-
-### GitHub Actions
-
-```bash
-npx instrlint init-ci --github
-```
-
-This creates `.github/workflows/instrlint.yml` with:
-- Triggers on changes to `CLAUDE.md`, `.claude/**`, `AGENTS.md`, etc.
-- Runs `instrlint ci --fail-on warning --format sarif`
-- Uploads SARIF to GitHub Code Scanning
-
-Or add it to your existing CI:
-
-```yaml
-- name: Run instrlint
-  run: npx instrlint@latest ci --fail-on warning
-```
+## Supported ecosystems (dead rule detection)
 
-### GitLab CI
+| Ecosystem | Config files checked |
+|-----------|----------------------|
+| JavaScript / TypeScript | `tsconfig.json`, `.eslintrc.*`, `.prettierrc.*`, `.editorconfig`, `commitlint.config.*` |
+| C# / .NET | `.csproj`, `Directory.Build.props`, `.editorconfig` (`dotnet_*`, `csharp_*` keys) |
 
-```bash
-npx instrlint init-ci --gitlab
-```
+More ecosystems (Python, Go, Rust) are planned. See [Contributing](#contributing) to add patterns.
 
-## Skill installation
+## Use in Claude Code or Codex
 
-Use instrlint directly from Claude Code or Codex without leaving the editor:
+Install the skill once:
 
 ```bash
-# Claude Code (global)
-npx instrlint install --claude-code
-
-# Claude Code (project only)
-npx instrlint install --claude-code --project
-
-# Codex
+npx instrlint install --claude-code          # global (~/.claude/commands/)
+npx instrlint install --claude-code --project  # project only
 npx instrlint install --codex
 ```
 
-Then **restart Claude Code** to activate the command. Then in your editor:
+Restart your editor to load the command, then:
 
 ```
-/instrlint
-/instrlint --fix
-/instrlint ci --fail-on warning
-/instrlint --verify
+/instrlint           # full report
+/instrlint --fix     # auto-fix safe issues
+/instrlint --verify  # LLM-verified report — no API key needed
 ```
 
 > **Note:** Claude Code only loads custom commands at startup. `/reload-plugins` does not pick up newly installed commands.
 
-### LLM verification (`/instrlint --verify`)
+The skill auto-detects when your report needs deeper review. `/instrlint --verify` triggers a two-pass protocol: instrlint writes low-confidence findings as `candidates.json`, the host model judges each one, and instrlint merges the verdicts — filtering false positives and attaching `✓` / `❓` badges. No API key; the model already running your session is the verifier.
+
+When your root file is too long, `/instrlint` walks you through a splitting decision: each section is classified as worth extracting (low load rate), extractable but always-loaded, should delete not move, or must stay. You choose the scope; the skill executes.
 
-The `--verify` flag triggers a two-pass protocol where the host agent (Claude Code or Codex) judges low-confidence findings to eliminate false positives:
+## Common commands
 
-1. **Emit candidates** — instrlint writes `instrlint-candidates.json` with contradiction/duplicate findings that need semantic review
-2. **Judge** — you (or the host agent) read the file and determine `confirmed` / `rejected` / `uncertain` for each
-3. **Write verdicts** — save decisions to `instrlint-verdicts.json`
-4. **Apply** — instrlint merges verdicts, filters rejected findings, and re-renders the report with `✓` / `❓` badges
+```bash
+instrlint                           # full health check
+instrlint --fix                     # auto-fix dead rules, stale refs, duplicates
+instrlint --format markdown         # Markdown output for PR comments
+instrlint --lang zh-TW              # Traditional Chinese output
 
-instrlint never calls an LLM API. It delegates judgment to whatever model is already running the session.
+instrlint ci --fail-on warning      # CI mode: exit 1 on warnings
+instrlint init-ci --github          # generate GitHub Actions workflow
 
-## Score and grade
+instrlint install --claude-code     # install as Claude Code skill
 
-| Grade | Score | Meaning |
-|-------|-------|---------|
-| A | 90–100 | Excellent |
-| B | 80–89 | Good |
-| C | 70–79 | Fair |
-| D | 60–69 | Poor |
-| F | < 60 | Critical |
+# LLM verification (two-pass, no API key)
+instrlint --emit-candidates instrlint-candidates.json
+instrlint --apply-verdicts instrlint-verdicts.json
+```
 
-Deductions: critical finding −10 (cap −40), warning −5 (cap −30), info −1 (cap −10). Root file length penalty: proportional above 200 lines (201–300: −5, 301–400: −8, 401–500: −10, 501–600: −15, 601+: −20, cap −30). Budget penalty: continuous above 25% of context window (cap −30).
+See all options: `instrlint --help`
 
 ## Contributing
 
@@ -196,7 +149,7 @@ The primary differentiator is the config overlap patterns in `src/detectors/conf
 
 1. Add a new entry to the `OVERLAP_PATTERNS` array
 2. Map the rule keywords to the config file + field that enforces them
-3. Add a test fixture in `tests/fixtures/sample-project/` that triggers the pattern
+3. Add a test fixture in `tests/fixtures/` that triggers the pattern
 4. Add a test case in `tests/detectors/config-overlap.test.ts`
 
 See [CLAUDE.md](CLAUDE.md) for full architecture documentation.

package/README.zh-TW.md

@@ -8,24 +8,33 @@
 
 **instrlint 會 lint 自己。**
 
-Lint 並優化你的 `CLAUDE.md` / `AGENTS.md` — 找出冗餘規則、token 浪費和結構問題。一個指令、一個分數、一份報告。
+## 為什麼需要 instrlint？
 
-## 快速開始
+`CLAUDE.md` 會隨時間膨脹。規則從範本複製過來、在不同段落重複出現、或留著指向早已不存在的檔案。有些規則默默地和 `tsconfig.json`、`.editorconfig` 重複——白白佔用 token 預算，什麼好處也沒有。
 
-**全域安裝（建議）：**
+instrlint 掃描你的指令檔，告訴你哪裡要修。一個指令、一個分數、一份可執行的報告。
 
-```bash
-npm install -g instrlint
-instrlint
-```
+有兩件事讓它和單純的行數計算工具不同：
 
-安裝一次，在任何專案都能直接使用。或用 `npx` 免安裝執行：
+- **Host 協調的 LLM 驗證** — 在 Claude Code 執行 `/instrlint --verify`，host model 自己判斷有疑義的 findings。不需要 API key，不需要外部服務。當前 session 裡已在運行的模型就是驗證者。
+- **拆分決策引導** — 根指令檔太長時，instrlint 不只說「去拆一拆」，而是逐段引導你做決定：每個段落會被分入四個桶，再依你選擇的範圍執行——因為並非每份過長的 CLAUDE.md 都該用同一種方式拆。
+
+支援 Claude Code、Codex 和 Cursor。487 個測試。從 v0.2.3 開始用自己 lint 自己。
+
+## 快速開始
 
 ```bash
 npx instrlint
 ```
 
-在含有 `CLAUDE.md` 或 `AGENTS.md` 的專案根目錄執行即可。
+在含有 `CLAUDE.md` 或 `AGENTS.md` 的專案根目錄執行即可，免安裝。
+
+或全域安裝，在任何專案都能直接使用：
+
+```bash
+npm install -g instrlint
+instrlint
+```
 
 ## 輸出範例
 
@@ -65,19 +74,17 @@ npx instrlint
 npx instrlint --lang zh-TW
 ```
 
-## 功能特色
+## 能找出什麼問題
 
-**Token 預算** — 使用 `cl100k_base` encoding 計算所有指令檔的 token 用量（無法載入時自動降級為字元估算）。標記超過 200 行的檔案，以及 baseline context 超過視窗 25% 的情況。
+**Token 預算** — 使用 `cl100k_base` encoding 計算所有指令檔和 MCP server 的 token 用量。標記超過 200 行的根指令檔，以及超過 context 視窗 25% 的情況。
 
-**冗餘規則偵測** — 偵測已被 config 檔強制執行的規則。涵蓋 TypeScript strict mode、Prettier 格式化、ESLint 規則、conventional commits、EditorConfig 等約 15 種重疊模式。
+**冗餘規則** — 偵測已被 config 檔強制執行的規則：TypeScript strict mode、Prettier 格式化、ESLint、EditorConfig、conventional commits、C# `.csproj`/`.editorconfig` 設定等，約 20 種模式，涵蓋 JS/TS 和 C# 生態系。
 
-**結構分析** — 找出互相矛盾的規則、過時的檔案參照、完全相同和高度相似的重複規則，以及路徑範圍化的優化機會。
+**結構問題** — 互相矛盾的規則、過時的檔案參照、高度相似的重複規則，以及路徑範圍化的優化機會。
 
 **自動修復** — `--fix` 安全地移除冗餘規則、過時參照和完全重複的規則。需要乾淨的 git 工作目錄。
 
-**LLM 驗證** — `/instrlint --verify`（Claude Code / Codex skill）執行兩段式驗證協議：instrlint 將低信心的 findings 輸出為 `candidates.json`，由你（host agent）逐一判斷後寫入 `verdicts.json`，再由 instrlint 合併結果 — 過濾 false positive，並附上 `✓ confirmed` / `❓ uncertain` badge。不需要任何 API key；host model 本身就是驗證者。
-
-**CI 整合** — `instrlint ci` 依 findings 嚴重程度回傳 exit code 0 或 1，並支援 SARIF 輸出供 GitHub Code Scanning 使用。
+**CI 整合** — `instrlint ci` 依嚴重程度回傳 exit code。支援 SARIF 輸出供 GitHub Code Scanning 使用。使用 `instrlint init-ci --github` 產生 workflow 設定。
 
 ## 支援工具
 
@@ -89,112 +96,58 @@ npx instrlint --lang zh-TW
 
 從專案結構自動偵測。可用 `--tool` 強制指定。
 
-## 指令
-
-```bash
-instrlint                         # 完整健康檢查 — 分數、等級、首要問題
-instrlint budget                  # 僅 token 預算分析
-instrlint deadrules               # 僅冗餘規則偵測
-instrlint structure               # 僅結構分析
-instrlint --fix                   # 自動修復安全問題
-instrlint --fix --force           # 跳過 git clean 檢查
-
-instrlint --format json           # JSON 輸出（供 CI、腳本使用）
-instrlint --format markdown       # Markdown 輸出（供 PR 留言使用）
-instrlint --lang zh-TW            # 繁體中文輸出
-
-instrlint ci                      # CI 模式：嚴重問題時 exit 1
-instrlint ci --fail-on warning    # 警告也 exit 1
-instrlint ci --format sarif       # SARIF 輸出供 GitHub Code Scanning
-instrlint ci --output report.sarif  # 寫入檔案
-
-instrlint init-ci --github        # 產生 .github/workflows/instrlint.yml
-instrlint init-ci --gitlab        # 輸出 GitLab CI 設定片段
-
-instrlint install --claude-code   # 安裝為全域 Claude Code skill
-instrlint install --claude-code --project  # 安裝到專案
-instrlint install --codex         # 安裝為 Codex skill
-
-# 由 host 協調的 LLM 驗證（兩段式，不需要 API key）
-instrlint --emit-candidates instrlint-candidates.json        # 將低信心 findings 輸出供 host LLM 判斷
-instrlint --emit-candidates instrlint-candidates.json --skip-report  # 同上，不印報告
-instrlint --apply-verdicts instrlint-verdicts.json           # 套用 host LLM 的判斷結果並重新呈現報告
-```
-
-## CI 整合
+## 支援生態系（冗餘規則偵測）
 
-### GitHub Actions
+| 生態系 | 檢查的 config 檔 |
+|--------|------------------|
+| JavaScript / TypeScript | `tsconfig.json`、`.eslintrc.*`、`.prettierrc.*`、`.editorconfig`、`commitlint.config.*` |
+| C# / .NET | `.csproj`、`Directory.Build.props`、`.editorconfig`（`dotnet_*`、`csharp_*` 欄位）|
 
-```bash
-npx instrlint init-ci --github
-```
+更多生態系（Python、Go、Rust）列入規劃。歡迎參閱[貢獻](#貢獻)新增規則。
 
-此指令會建立 `.github/workflows/instrlint.yml`，包含：
-- 在 `CLAUDE.md`、`.claude/**`、`AGENTS.md` 等檔案變更時觸發
-- 執行 `instrlint ci --fail-on warning --format sarif`
-- 上傳 SARIF 至 GitHub Code Scanning
+## 在 Claude Code 或 Codex 中使用
 
-或加入現有 CI：
-
-```yaml
-- name: Run instrlint
-  run: npx instrlint@latest ci --fail-on warning
-```
-
-### GitLab CI
+安裝一次 skill：
 
 ```bash
-npx instrlint init-ci --gitlab
-```
-
-## Skill 安裝
-
-不離開編輯器直接從 Claude Code 或 Codex 使用 instrlint：
-
-```bash
-# Claude Code（全域）
-npx instrlint install --claude-code
-
-# Claude Code（僅專案）
-npx instrlint install --claude-code --project
-
-# Codex
+npx instrlint install --claude-code          # 全域（~/.claude/commands/）
+npx instrlint install --claude-code --project  # 僅專案
 npx instrlint install --codex
 ```
 
-安裝後**重新啟動 Claude Code** 以啟用指令，然後在編輯器中使用：
+重新啟動編輯器以載入指令，然後使用：
 
 ```
-/instrlint
-/instrlint --fix
-/instrlint ci --fail-on warning
-/instrlint --verify
+/instrlint           # 完整報告
+/instrlint --fix     # 自動修復安全問題
+/instrlint --verify  # LLM 驗證報告 — 不需要 API key
 ```
 
 > **注意：** Claude Code 只在啟動時載入 custom commands。`/reload-plugins` 無法載入新安裝的指令。
 
-### LLM 驗證（`/instrlint --verify`）
+當報告需要更深入的審查，skill 會自動觸發兩段式驗證：instrlint 將低信心的 findings 寫成 `candidates.json`，host model 逐一判斷後寫 `verdicts.json`，再由 instrlint 合併結果——過濾 false positive，並附上 `✓` / `❓` badge。不需要 API key；當前 session 的模型就是驗證者。
+
+根指令檔過長時，`/instrlint` 會引導你做拆分決策：每個段落被分入四個桶——值得抽出（低載入率）、可抽出但不省 token、該刪不是該搬、或必須留在原檔。你選範圍，skill 執行。
 
-`--verify` 觸發兩段式協議，由 host agent（Claude Code 或 Codex）判斷低信心的 findings 以消除 false positive：
+## 常用指令
 
-1. **輸出 candidates** — instrlint 將需要語意審查的 contradiction / duplicate findings 寫成 `instrlint-candidates.json`
-2. **判斷** — 你（或 host agent）讀取檔案，對每筆 finding 決定 `confirmed` / `rejected` / `uncertain`
-3. **寫 verdicts** — 將判斷結果儲存為 `instrlint-verdicts.json`
-4. **套用** — instrlint 合併 verdicts，過濾被 rejected 的 findings，並重新輸出附有 `✓` / `❓` badge 的報告
+```bash
+instrlint                           # 完整健康檢查
+instrlint --fix                     # 自動修復冗餘規則、過時參照、重複規則
+instrlint --format markdown         # Markdown 輸出（供 PR 留言使用）
+instrlint --lang zh-TW              # 繁體中文輸出
 
-instrlint 從不呼叫任何 LLM API，而是將判斷委託給當前 session 中正在運行的 model。
+instrlint ci --fail-on warning      # CI 模式：警告也 exit 1
+instrlint init-ci --github          # 產生 GitHub Actions workflow
 
-## 分數與等級
+instrlint install --claude-code     # 安裝為 Claude Code skill
 
-| 等級 | 分數 | 說明 |
-|------|------|------|
-| A | 90–100 | 優秀 |
-| B | 80–89 | 良好 |
-| C | 70–79 | 尚可 |
-| D | 60–69 | 較差 |
-| F | < 60 | 嚴重 |
+# LLM 驗證（兩段式，不需要 API key）
+instrlint --emit-candidates instrlint-candidates.json
+instrlint --apply-verdicts instrlint-verdicts.json
+```
 
-扣分規則：嚴重問題 −10（上限 −40）、警告 −5（上限 −30）、建議 −1（上限 −10）。根指令檔行數懲罰：超過 200 行後按比例扣分（201–300 行 −5、301–400 −8、401–500 −10、501–600 −15、601+ −20，上限 −30）。預算懲罰：超過 context 視窗 25% 後連續計算（上限 −30）。
+所有選項請執行：`instrlint --help`
 
 ## 貢獻
 
@@ -202,7 +155,7 @@ instrlint 從不呼叫任何 LLM API，而是將判斷委託給當前 session
 
 1. 在 `OVERLAP_PATTERNS` 陣列新增一筆
 2. 將規則關鍵字對應到強制執行它的 config 檔 + 欄位
-3. 在 `tests/fixtures/sample-project/` 新增觸發該規則的 fixture
+3. 在 `tests/fixtures/` 新增觸發該規則的 fixture
 4. 在 `tests/detectors/config-overlap.test.ts` 新增測試案例
 
 完整架構文件請參閱 [CLAUDE.md](CLAUDE.md)。

package/dist/cli.cjs

@@ -119,11 +119,21 @@ function scanProject(projectRoot, forceTool) {
     if (byFile != null) {
       return { ...byFile, confidence: "low" };
     }
-    return { tool: "unknown", rootFilePath: null, configDir: null, confidence: "low" };
+    return {
+      tool: "unknown",
+      rootFilePath: null,
+      configDir: null,
+      confidence: "low"
+    };
   }
   if (detections.length === 1) {
     const d = detections[0];
-    return { tool: d.tool, rootFilePath: d.rootFilePath, configDir: d.configDir, confidence: "high" };
+    return {
+      tool: d.tool,
+      rootFilePath: d.rootFilePath,
+      configDir: d.configDir,
+      confidence: "high"
+    };
   }
   const first = detections[0];
   return {
@@ -2511,7 +2521,9 @@ function printStructureSuggestions(suggestions, projectRoot, output) {
     } else {
       const dir = s.pathDir ?? "src";
       const { filePath, content } = buildPathScopedFile(dir, s.ruleText);
-      output.log(`  ${import_chalk3.default.cyan(t("fix.pathScopedCreate", { path: filePath }))}`);
+      output.log(
+        `  ${import_chalk3.default.cyan(t("fix.pathScopedCreate", { path: filePath }))}`
+      );
       terminalCodeBlock(content, output);
       if (lineNum > 0) {
         output.log(
@@ -2534,7 +2546,14 @@ function markdownStructureSuggestions(suggestions, projectRoot) {
       ""
     );
     if (s.type === "hook") {
-      lines.push(t("fix.hookCreate"), "", "```json", buildHookSnippet(s.ruleText), "```", "");
+      lines.push(
+        t("fix.hookCreate"),
+        "",
+        "```json",
+        buildHookSnippet(s.ruleText),
+        "```",
+        ""
+      );
       lines.push(`> ${t("fix.hookWarning")}`, "");
       if (lineNum > 0) {
         lines.push(
@@ -2842,18 +2861,30 @@ var import_fs12 = require("fs");
 var import_path10 = require("path");
 var import_os2 = require("os");
 var import_url2 = require("url");
-function resolveSkillFile(target) {
+function resolveSkillFile() {
   const thisFile = (0, import_url2.fileURLToPath)(importMetaUrl);
-  const subDir = target === "claude-code" ? "claude-code" : "codex";
-  for (const levels of [2, 3]) {
+  const levelCandidates = [2, 3];
+  for (const levels of levelCandidates) {
     const parts = Array(levels).fill("..");
-    const candidate = (0, import_path10.join)(thisFile, ...parts, "skills", subDir, "SKILL.md");
+    const candidate = (0, import_path10.join)(
+      thisFile,
+      ...parts,
+      "skills",
+      "instrlint",
+      "SKILL.md"
+    );
     if ((0, import_fs12.existsSync)(candidate)) return candidate;
   }
-  return (0, import_path10.join)(thisFile, "..", "..", "skills", subDir, "SKILL.md");
+  return (0, import_path10.join)(
+    thisFile,
+    ...Array(levelCandidates[0]).fill(".."),
+    "skills",
+    "instrlint",
+    "SKILL.md"
+  );
 }
-function readSkillContent(target) {
-  const skillPath = resolveSkillFile(target);
+function readSkillContent() {
+  const skillPath = resolveSkillFile();
   try {
     const raw = (0, import_fs12.readFileSync)(skillPath, "utf8");
     return injectVersion(raw, CURRENT_VERSION);
@@ -2890,14 +2921,14 @@ function installCodex(content, projectRoot, force, output) {
 function runInstall(opts, output = console) {
   const projectRoot = opts.projectRoot ?? process.cwd();
   const force = opts.force ?? false;
+  let content;
+  try {
+    content = readSkillContent();
+  } catch (err) {
+    output.error(String(err));
+    return { exitCode: 1, errorMessage: String(err) };
+  }
   if (opts.claudeCode) {
-    let content;
-    try {
-      content = readSkillContent("claude-code");
-    } catch (err) {
-      output.error(String(err));
-      return { exitCode: 1, errorMessage: String(err) };
-    }
     return installClaudeCode(
       content,
       projectRoot,
@@ -2907,13 +2938,6 @@ function runInstall(opts, output = console) {
     );
   }
   if (opts.codex) {
-    let content;
-    try {
-      content = readSkillContent("codex");
-    } catch (err) {
-      output.error(String(err));
-      return { exitCode: 1, errorMessage: String(err) };
-    }
     return installCodex(content, projectRoot, force, output);
   }
   output.error(t("install.unknownTarget"));