@yeongjaeyou/claude-code-config 0.10.0 → 0.11.0

package/.claude/commands/edit-notebook.md

@@ -34,6 +34,14 @@ Safely edit Jupyter Notebook files using the correct tools.
    - Check for missing functions/imports/dependencies
    - Preserve cell outputs unless explicitly instructed otherwise
 
+5. **Key Insights verification**: Verify outputs before documenting
+   - **Required steps**:
+     1. Read notebook output cells (use Read tool)
+     2. Read generated data files (CSV, JSON, etc.)
+     3. Check visualizations/images if any
+   - **Never**: Write insights based only on code flow
+   - **Required**: Document data source (e.g., "Source: outputs/analysis.csv")
+
 ## Guidelines
 
 - **Use dedicated tool only**: Never use editing tools other than NotebookEdit
@@ -41,3 +49,15 @@ Safely edit Jupyter Notebook files using the correct tools.
 - **Verification required**: Immediately verify changes after editing
 - **Follow CLAUDE.md**: Adhere strictly to project guidelines
 
+## Computation Efficiency
+
+### Avoid
+- Repeated model loading across cells
+- Re-running inference when results exist
+- Computing without caching
+
+### Required patterns
+1. **Cache-first**: Load cached results if exist, compute only if missing
+2. **Single instance**: Load model once in early cell, reuse throughout
+3. **Filter over recompute**: Filter cached results instead of re-running with different parameters
+

package/.claude/commands/gh/decompose-issue.md

@@ -5,16 +5,20 @@ Break down large work items into manageable, independent issues. Follow project
 ## Workflow
 
 1. Check issue numbers: Run `gh issue list` to view current issue numbers
-2. Analyze work: Understand core requirements and objectives
-3. Decompose work: Split major tasks into smaller, manageable sub-tasks or issues. **Aim for optimal count over excessive issues (keep it manageable)**
-4. Analyze dependencies: Identify prerequisite tasks
-5. Suggest milestone name: Propose a milestone to group decomposed tasks
-6. Check related PRs (optional): Run `gh pr list --state closed --limit 20` for similar work references (skip if none)
-7. Output decomposed issues: Display issues with proposed milestone name
-8. Ask about GitHub creation: Use AskUserQuestion to let user decide on milestone and issue creation
+2. **Discover available components**:
+   - Scan `.claude/agents/` for custom agents (read YAML frontmatter)
+   - Scan `.claude/skills/` for available skills (read SKILL.md)
+   - Check `.mcp.json` for configured MCP servers
+3. Analyze work: Understand core requirements and objectives
+4. Decompose work: Split major tasks into smaller, manageable sub-tasks or issues. **Aim for optimal count over excessive issues (keep it manageable)**
+5. Analyze dependencies: Identify prerequisite tasks
+6. Suggest milestone name: Propose a milestone to group decomposed tasks
+7. Check related PRs (optional): Run `gh pr list --state closed --limit 20` for similar work references (skip if none)
+8. Output decomposed issues: Display issues with proposed milestone name
+9. Ask about GitHub creation: Use AskUserQuestion to let user decide on milestone and issue creation
    - Create milestone: `gh api repos/:owner/:repo/milestones -f title="Milestone Name" -f description="Description"`
    - Assign issues with `--milestone` option
-9. **Add issues to GitHub Project (optional)**
+10. **Add issues to GitHub Project (optional)**
    - Check for existing projects: `gh project list --owner <owner> --format json`
    - If no project exists: Display "No project found. You can create one with `/gh:init-project`" and skip
    - If project exists: Ask user via AskUserQuestion whether to add issues
@@ -58,9 +62,99 @@ Examples (vary by project, for reference only):
 **Dependencies**:
 - [ ] None or prerequisite issue #number
 
-**Recommended agent** (if applicable):
-- Include here if agent usage is specified in arguments
+**Execution strategy**:
+- **Pattern**: main-only | sequential | parallel | delegation
+- **Delegate to**: (delegation only) agent name
+- **Skills**: [discovered skills, or none]
+- **MCP**: [discovered MCP servers, or none]
+
+**Execution diagram**:
+```
++------+
+| main |
++------+
+Skills: none
+MCP: none
+```
 
 **References** (optional):
 - Add related PRs if available (e.g., PR #36 - brief description)
 - Omit this section if none
+
+---
+
+## Execution Strategy Guide
+
+### Component Discovery
+
+Before decomposing issues, scan for available components:
+- **Agents**: `.claude/agents/*.md` (extract name, description from YAML frontmatter)
+- **Skills**: `.claude/skills/*/SKILL.md` (extract name, description from YAML frontmatter)
+- **MCP**: `.mcp.json` (extract server names and purposes)
+
+### Execution Patterns
+
+```
+Pattern A: main-only
++------+ Skills: /code-explorer, /xlsx
+| main | MCP: serena, context7
++------+
+
+Pattern B: sequential
++---------+     +------+ Skills: ...
+| explore | --> | main | MCP: ...
++---------+     +------+
+
+Pattern C: parallel
++---------+
+| explore |--+
++---------+  |  +------+ Skills: ...
+| explore |--+->| main | MCP: ...
++---------+  |  +------+
+| explore |--+
++---------+
+
+Pattern D: delegation
++------+     +----------------+ Skills: ...
+| main | --> | python-pro     | MCP: serena
++------+     | web-researcher |
+             +----------------+
+
+Note: main can spawn subagents at any point during execution
+```
+
+### Pattern Selection
+
+| Situation | Pattern | Example |
+|-----------|---------|---------|
+| Simple fix/change | main-only | Bug fix, single file change |
+| Analysis then implement | sequential | New feature, refactoring |
+| Complex analysis needed | parallel | Large refactor, architecture change |
+| Specialized work | delegation | Python optimization, technical research |
+
+### Diagram in Issues
+
+Always include an ASCII diagram in the issue body to visualize the execution flow:
+
+```
+// main-only
++------+
+| main |
++------+
+Skills: /code-review
+MCP: serena
+
+// delegation
++------+     +--------------------+
+| main | --> | deepfake-cv-expert |
++------+     +--------------------+
+Skills: none
+MCP: none
+
+// sequential
++---------+     +------+
+| explore | --> | main |
++---------+     +------+
+Skills: ...
+MCP: ...
+```

package/.claude/guidelines/work-guidelines.md

@@ -61,6 +61,10 @@ Utilize MCP servers whenever possible:
 - `mcpdocs` - Documentation fetching
 - `firecrawl` - Web scraping and search
 
+### Agent & Skill Usage
+- 작업에 적합한 agent 또는 skill이 있다면 적극 활용
+- 의도가 명확하면 바로 사용, 불명확하면 제안 후 승인 받기
+
 ### Code Exploration Strategy
 
 #### Parallel Exploration with Multiple Tools

package/.claude/hooks/inject-guidelines.sh

@@ -0,0 +1,37 @@
+#!/bin/bash
+# UserPromptSubmit Hook: Auto-inject guidelines on every prompt
+# Location: .claude/hooks/inject-guidelines.sh
+
+set -e
+
+# Project directory (provided by Claude Code)
+PROJECT_DIR="${CLAUDE_PROJECT_DIR:-$(pwd)}"
+GUIDELINES_DIR="$PROJECT_DIR/.claude/guidelines"
+
+# Primary guideline file to always inject
+PRIMARY_GUIDELINE="$GUIDELINES_DIR/work-guidelines.md"
+
+# Output as system-reminder for Claude to process
+if [ -f "$PRIMARY_GUIDELINE" ]; then
+  echo "<system-reminder>"
+  echo "Called the Read tool with the following input: {\"file_path\":\"$PRIMARY_GUIDELINE\"}"
+  echo "</system-reminder>"
+  echo "<system-reminder>"
+  echo "Result of calling the Read tool:"
+  cat "$PRIMARY_GUIDELINE"
+  echo "</system-reminder>"
+fi
+
+# Uncomment below to inject ALL guidelines in the directory
+# if [ -d "$GUIDELINES_DIR" ]; then
+#   for file in "$GUIDELINES_DIR"/*.md; do
+#     if [ -f "$file" ]; then
+#       echo "<system-reminder>"
+#       echo "### $(basename "$file")"
+#       cat "$file"
+#       echo "</system-reminder>"
+#     fi
+#   done
+# fi
+
+exit 0

package/bin/cli.js

@@ -9,7 +9,7 @@ const pkg = require('../package.json');
 const args = process.argv.slice(2);
 
 // Installation target folders (non-user data only)
-const INSTALL_FOLDERS = ['commands', 'agents', 'skills', 'guidelines'];
+const INSTALL_FOLDERS = ['commands', 'agents', 'skills', 'guidelines', 'hooks'];
 
 // Parse options
 const isGlobal = args.includes('--global') || args.includes('-g');
@@ -44,6 +44,7 @@ Installed folders:
   - agents/     : Custom agents
   - skills/     : Skills (reusable tool collections)
   - guidelines/ : Shared guidelines
+  - hooks/      : Hook scripts (auto-inject guidelines)
 
 Note:
   With --global install, existing user data (settings.json, history.jsonl, etc.) is preserved.
@@ -133,6 +134,10 @@ function copyFolder(src, dst, mode = 'merge') {
       // Custom files (only in dest) are preserved since we don't delete them
       try {
         fs.copyFileSync(srcPath, dstPath);
+        // Set execute permission for shell scripts
+        if (entry.name.endsWith('.sh')) {
+          fs.chmodSync(dstPath, 0o755);
+        }
       } catch (err) {
         throw new Error(`Failed to copy file: ${srcPath} -> ${dstPath} - ${err.message}`);
       }
@@ -160,6 +165,81 @@ function installFolders(mode) {
   }
 }
 
+// Default hook configuration for inject-guidelines
+const DEFAULT_HOOKS_CONFIG = {
+  UserPromptSubmit: [
+    {
+      matcher: '',
+      hooks: [
+        {
+          type: 'command',
+          command: 'bash .claude/hooks/inject-guidelines.sh'
+        }
+      ]
+    }
+  ]
+};
+
+/**
+ * Check if a hook with the same command already exists
+ * @param {Array} hooks - Existing hooks array
+ * @param {string} command - Command to check
+ * @returns {boolean}
+ */
+function hookExists(hooks, command) {
+  if (!Array.isArray(hooks)) return false;
+  return hooks.some(hook => {
+    if (!hook.hooks || !Array.isArray(hook.hooks)) return false;
+    return hook.hooks.some(h => h.command === command);
+  });
+}
+
+/**
+ * Merge settings.json with default hooks configuration
+ * Preserves existing user settings and only adds new hooks
+ */
+function mergeSettingsJson() {
+  const settingsPath = path.join(dest, 'settings.json');
+  let settings = {};
+
+  // Load existing settings if exists
+  if (fs.existsSync(settingsPath)) {
+    try {
+      const content = fs.readFileSync(settingsPath, 'utf8');
+      settings = JSON.parse(content);
+    } catch (err) {
+      console.log('Warning: Could not parse existing settings.json, creating new one.');
+      settings = {};
+    }
+  }
+
+  // Initialize hooks object if not exists
+  if (!settings.hooks) {
+    settings.hooks = {};
+  }
+
+  // Initialize UserPromptSubmit array if not exists
+  if (!settings.hooks.UserPromptSubmit) {
+    settings.hooks.UserPromptSubmit = [];
+  }
+
+  // Add default hooks if not already present
+  const targetCommand = DEFAULT_HOOKS_CONFIG.UserPromptSubmit[0].hooks[0].command;
+  if (!hookExists(settings.hooks.UserPromptSubmit, targetCommand)) {
+    settings.hooks.UserPromptSubmit.push(DEFAULT_HOOKS_CONFIG.UserPromptSubmit[0]);
+    console.log('Added inject-guidelines hook to settings.json');
+  } else {
+    console.log('inject-guidelines hook already exists in settings.json');
+  }
+
+  // Write merged settings
+  try {
+    fs.writeFileSync(settingsPath, JSON.stringify(settings, null, 2) + '\n');
+  } catch (err) {
+    throw new Error(`Failed to write settings.json: ${err.message}`);
+  }
+}
+
 /**
  * Main function
  */
@@ -237,6 +317,9 @@ async function main() {
   // Install folders
   installFolders(installMode);
 
+  // Merge settings.json with hooks configuration
+  mergeSettingsJson();
+
   console.log('');
   console.log('.claude/ folder installed successfully!');
   console.log('');
@@ -245,6 +328,10 @@ async function main() {
   console.log('  - agents/     : Custom agents');
   console.log('  - skills/     : Skills (reusable tool collections)');
   console.log('  - guidelines/ : Shared guidelines');
+  console.log('  - hooks/      : Hook scripts (auto-inject guidelines)');
+  console.log('');
+  console.log('Configured:');
+  console.log('  - settings.json : UserPromptSubmit hook registered');
   console.log('');
 
   if (isGlobal) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yeongjaeyou/claude-code-config",
-  "version": "0.10.0",
+  "version": "0.11.0",
   "description": "Claude Code CLI custom commands, agents, and skills",
   "bin": {
     "claude-code-config": "./bin/cli.js"