cokit-cli 1.0.7 → 1.0.9

package/plans/260106-1102-cokit-implementation/plan.md

@@ -1,130 +0,0 @@
----
-title: "CoKit: Claude Code to GitHub Copilot Port"
-description: "Full port of Claude Code workflow patterns to GitHub Copilot with easy CLI"
-status: completed
-priority: P1
-effort: 12h
-branch: main
-tags: [cokit, copilot, port, cli, npm]
-created: 2026-01-06
-completed_phases: [1, 2, 3, 4, 5]
-phase3_completed: 2026-01-06
-phase4_completed: 2026-01-06
-phase5_completed: 2026-01-06
----
-
-# CoKit Implementation Plan
-
-## Overview
-
-Port Claude Code patterns to GitHub Copilot with **dead-simple CLI**:
-```bash
-npx cokit init          # Setup in current project
-npx cokit init --global # Setup personal skills
-```
-
-## User Experience Priority
-
-Target: **Non-technical users** who want AI workflow without complexity.
-- One command to install
-- Simple README with screenshots
-- No config files to edit manually
-
-## Architecture
-
-```
-cokit/
-├── package.json                  # npm package config
-├── bin/
-│   └── cokit.js                  # CLI entry point
-├── src/
-│   ├── cli.js                    # Command parser
-│   ├── commands/
-│   │   ├── init.js               # npx cokit init
-│   │   ├── add.js                # npx cokit add <skill>
-│   │   ├── list.js               # npx cokit list
-│   │   └── update.js             # npx cokit update
-│   └── utils/
-│       ├── copy.js               # File operations
-│       └── prompt.js             # User prompts
-├── templates/
-│   └── repo/                     # Repo-level templates
-│       ├── .github/
-│       │   ├── copilot-instructions.md
-│       │   ├── AGENTS.md
-│       │   ├── instructions/
-│       │   ├── prompts/
-│       │   └── skills/
-│       └── .vscode/
-│           └── settings.json     # VS Code Copilot config
-├── skills/                       # User-level skills
-│   ├── debugging/
-│   ├── code-review/
-│   ├── planning/
-│   ├── docs-seeker/
-│   └── sequential-thinking/
-├── README.md                     # Simple getting started
-└── LICENSE
-```
-
-## Key Mappings
-
-| Claude Code | Copilot | Location |
-|-------------|---------|----------|
-| CLAUDE.md | copilot-instructions.md | `.github/` |
-| skills/ | skills/ | `.github/skills/` or `~/.copilot/skills/` |
-| agents/*.md | AGENTS.md | `.github/` (merged sections) |
-| commands/*.md | prompts/*.prompt.md | `.github/prompts/` |
-| workflows/*.md | instructions/*.instructions.md | `.github/instructions/` |
-| hooks | N/A | Skip - no equivalent |
-
-## Limitations (From Research)
-
-- **No hooks:** Static config only, no runtime injection
-- **No multi-agent:** Single agent, manual prompt chaining
-- **No $ARGUMENTS:** User provides context in chat
-- **No session state:** Stateless between interactions
-
-## Phases
-
-| Phase | Description | Effort | Link |
-|-------|-------------|--------|------|
-| 1 | CLI tool (npm package) | 4h | [phase-01](./phases/phase-01-cli-tool.md) |
-| 2 | Repo templates | 3h | [phase-02](./phases/phase-02-repo-level-templates.md) |
-| 3 | User skills (5 core) | 4h | [phase-03](./phases/phase-03-user-level-skills.md) |
-| 4 | Prompt files (6 core) | 2h | [phase-04](./phases/phase-04-prompt-files.md) |
-| 5 | Simple docs + README | 3h | [phase-05](./phases/phase-05-documentation.md) |
-
-## CLI Commands
-
-```bash
-npx cokit init              # Init repo templates (.github/)
-npx cokit init --global     # Init user skills (~/.copilot/skills/)
-npx cokit init --all        # Both repo + user
-npx cokit add <skill>       # Add specific skill
-npx cokit list              # Show installed skills/prompts
-npx cokit update            # Update to latest version
-npx cokit doctor            # Check setup + troubleshoot
-```
-
-## Success Criteria
-
-- [ ] `npx cokit init` works without prior install
-- [ ] Interactive prompts guide non-technical users
-- [ ] README has screenshots and GIFs
-- [ ] All 5 core skills ported
-- [ ] All 6 core prompts ported
-- [ ] VS Code settings template included
-- [ ] `cokit doctor` diagnoses common issues
-
-## Dependencies
-
-- GitHub Copilot Agent Mode (Feb 2025+)
-- VS Code with Copilot extension
-- Agent Skills support (Dec 2025+)
-
-## Unresolved Questions
-
-1. Does `~/.copilot/skills/` discovery work in JetBrains IDEs?
-2. Can skills reference scripts with relative paths across platforms?
-3. How does Copilot handle conflicting skills at repo vs user level?

package/plans/260106-1102-cokit-implementation/research/researcher-01-copilot-skills-spec.md

@@ -1,182 +0,0 @@
-# Research Report: GitHub Copilot Agent Skills Specification
-
-**Date:** January 6, 2026 | **Version:** 1.0
-
-## Executive Summary
-
-GitHub Copilot Agent Skills is a modular, extensible system for enhancing Claude Code (Anthropic's agentic coding tool) with specialized capabilities. Skills are self-contained packages containing a required `SKILL.md` file with YAML frontmatter + Markdown instructions, plus optional bundled resources (scripts, references, assets). Skills enable progressive disclosure of context, auto-discovery via semantic matching, and fine-grained tool access control via `allowed-tools`. The specification launched October 2025 (v1.0) and currently supports basic resource bundling with emerging support for metadata-driven customization.
-
-## Key Findings
-
-### 1. SKILL.md File Format
-
-**Required structure:** YAML frontmatter + Markdown body
-
-**Frontmatter (required fields):**
-- `name` - Hyphen-case, lowercase alphanumeric + hyphens, must match directory name
-- `description` - Explains purpose and activation triggers for Claude
-
-**Frontmatter (optional fields):**
-- `license` - License identifier (e.g., "MIT") or file reference (e.g., "LICENSE.txt")
-- `version` - Semantic version string (e.g., "1.0.0")
-- `allowed-tools` - Array of pre-approved tool names (Bash, Read, Write, Edit, etc.) for execution control
-- `metadata` - String key-value map for client-specific custom properties
-
-**Markdown body:** Unrestricted content, typically structured as:
-- Feature descriptions and use cases
-- Step-by-step workflows
-- Quick reference guides
-- References to bundled resources
-- Best practices and common patterns
-
-**Size constraint:** <100 lines recommended (leverage progressive disclosure by splitting into multiple reference files)
-
-### 2. Skill Discovery Mechanism
-
-**Auto-activation trigger:**
-- Claude Code analyzes user query intent semantically
-- Matches intent against skill descriptions and bundled resource metadata
-- Activates multiple skills if needed to fulfill user request
-- Skills load progressively (only referenced resources pulled into context)
-
-**Directory structure for discovery:**
-- User-level: `$HOME/.claude/skills/{skill-name}/SKILL.md` (automatic discovery)
-- Repo-level: Not natively supported (must use hooks/configuration workarounds)
-
-**Activation criteria:**
-- Description must contain concrete use cases/keywords matching user query
-- Descriptions should be concise but rich in actionable trigger phrases
-- Example: "When building mobile apps, implementing mobile UX patterns, optimizing for mobile constraints..."
-
-### 3. Resource Bundling & Progressive Disclosure
-
-**Optional subdirectories inside skill folder:**
-
-| Directory | Purpose | Example |
-|-----------|---------|---------|
-| `scripts/` | Executable code (Node.js/Python preferred over Bash) | analyze.js, fetch-docs.js, utils/ |
-| `references/` | Documentation loaded contextually as needed | patterns.md, errors.md, workflows.md |
-| `assets/` | Static files used in output | templates, icons, fonts |
-
-**Progressive disclosure principle:**
-- SKILL.md acts as index/entry point only (<100 lines)
-- Referenced files loaded only when Claude selects them based on user need
-- Reduces token consumption compared to monolithic documentation
-- Enables deep expertise without bloating initial context
-
-**Environment variable precedence for scripts:**
-1. Process environment variables
-2. `$HOME/.claude/skills/{skill-name}/.env`
-3. `$HOME/.claude/skills/.env`
-4. `$HOME/.claude/.env`
-
-### 4. Tool Access Control (allowed-tools)
-
-**Purpose:** Pre-approve specific tools for skill execution to prevent tool-related errors
-
-**Supported values (examples):** Bash, Read, Write, Edit, Grep, WebSearch, WebFetch
-
-**Behavior:**
-- If `allowed-tools` specified, only listed tools execute within skill context
-- Unapproved tools are skipped/blocked with runtime warnings
-- Improves reliability and security posture
-
-**Real-world example (ai-multimodal skill):**
-```yaml
-allowed-tools:
-  - Bash
-  - Read
-  - Write
-  - Edit
-```
-
-### 5. User-Level vs. Repo-Level Locations
-
-**User-level skills:**
-- Located: `$HOME/.claude/skills/` (e.g., `/Users/admin/.claude/skills/`)
-- Scope: Available to all projects
-- Discovery: Automatic by Claude Code
-- Use case: General-purpose domain expertise, common patterns
-
-**Repo-level skills:**
-- Not natively supported by Agent Skills Spec
-- Workaround: Use hooks (SessionStart, PreToolUse) to inject skill content or symlink from ~/.claude/skills/
-- Planned feature: future specification updates may add native repo-level skill paths
-
-### 6. Limitations & Known Issues
-
-**Current constraints:**
-- Skill discovery is semantic (no keyword-based exact matching)
-- Repo-level skills require manual configuration via hooks
-- No built-in skill versioning or dependency management
-- YAML frontmatter strict typing (no nested objects, only strings + arrays)
-- No native skill marketplace/registry integration (requires custom distribution)
-- Scripts must handle their own dependency management (no global package registry)
-- Tool access control (`allowed-tools`) relatively new, edge cases possible
-
-**Token consumption:**
-- While progressive disclosure reduces overhead, deeply nested references still require careful planning
-- Large scripts still consume tokens if loaded into context
-
-**Platform considerations:**
-- Bash scripts discouraged for cross-platform (Windows) compatibility
-- Python/Node.js scripts preferred with `requirements.txt` or `package.json`
-
-## Best Practices
-
-1. **Description depth:** Include concrete use cases so semantic matching activates correctly
-2. **File organization:** Keep SKILL.md <100 lines; split references into focused .md files <100 lines each
-3. **Scripts:** Use Node.js or Python; include `.env.example`; write unit tests
-4. **Resource references:** Use links/paths in SKILL.md to guide Claude to relevant bundled files
-5. **License clarity:** Always specify; recommend MIT for open skills
-6. **Versioning:** Use semantic versioning in optional `version` field for tracking updates
-
-## Specification Details
-
-**Version:** 1.0 (Public Launch October 16, 2025)
-
-**Minimal skill example:**
-```
-my-skill/
-  ├── SKILL.md
-```
-
-**Complex skill example (docs-seeker):**
-```
-docs-seeker/
-  ├── SKILL.md
-  ├── package.json
-  ├── scripts/
-  │   ├── analyze-llms-txt.js
-  │   ├── detect-topic.js
-  │   ├── fetch-docs.js
-  │   ├── tests/
-  │   └── utils/
-  ├── references/
-  │   ├── advanced.md
-  │   ├── context7-patterns.md
-  │   └── errors.md
-  └── workflows/
-      ├── library-search.md
-      ├── repo-analysis.md
-      └── topic-search.md
-```
-
-## Sources
-
-- **Agent Skills Spec** (Official): `/Users/admin/.claude/skills/agent_skills_spec.md` (v1.0, Oct 2025)
-- **Real-world implementations:**
-  - Research skill: `/Users/admin/.claude/skills/research/SKILL.md`
-  - Backend Development: `/Users/admin/.claude/skills/backend-development/SKILL.md`
-  - AI Multimodal: `/Users/admin/.claude/skills/ai-multimodal/SKILL.md`
-  - Skill Creator: `/Users/admin/.claude/skills/skill-creator/SKILL.md`
-  - Docs Seeker: `/Users/admin/.claude/skills/docs-seeker/` (complex structure)
-  - Claude Code: `/Users/admin/.claude/skills/claude-code/SKILL.md` (references other skills)
-
-## Unresolved Questions
-
-1. What is the exact semantic matching algorithm Claude uses for skill activation?
-2. Is there a planned timeline for native repo-level skill support?
-3. How are skill conflicts handled when multiple skills describe identical functionality?
-4. What is the max recommended size for bundled scripts before performance degradation?
-5. Can skills be nested (skill referencing another skill)?

package/plans/260106-1102-cokit-implementation/research/researcher-02-copilot-instructions-prompts.md

@@ -1,201 +0,0 @@
-# Research Report: GitHub Copilot Custom Instructions & Prompt Files
-
-**Conducted:** 2026-01-06 | **Sources:** 10+ GitHub Official Docs & Blog Articles
-
-## Executive Summary
-
-GitHub Copilot supports 4 primary customization file types: `copilot-instructions.md` (repo-wide), `*.instructions.md` (scoped), `*.prompt.md` (reusable tasks), and `AGENTS.md` (nested precedence). All use Markdown + YAML frontmatter. Key innovation: path-scoped instructions w/ `applyTo` glob patterns. AGENTS.md now officially standardized under Linux Foundation (Aug 2025). File precedence: nested > root > explicit user input overrides all.
-
-## 1. copilot-instructions.md
-
-**Location:** `.github/copilot-instructions.md` (workspace root)
-**Scope:** Repository-wide, applies to all Copilot agents & code review
-**Format:** Plain Markdown (no frontmatter required)
-**Use Case:** General coding standards, team guidelines, deprecated libraries, architectural patterns
-
-**Best Practice:**
-- Reserve for team-wide conventions
-- Keep concise & focused
-- Include preferred frameworks, testing requirements, security considerations
-
-## 2. *.instructions.md Files
-
-**Location:** `.github/instructions/` folder
-**Scope:** Path-specific via `applyTo` glob patterns
-**Format:**
-```yaml
----
-applyTo: "src/api/**/*.ts"
-excludeAgent: "code-review"  # Optional: "coding-agent" or "code-review"
-description: "API layer guidelines"
----
-# Your instructions
-```
-
-**applyTo Patterns:**
-- Glob syntax supported (e.g., `**/*.py`, `app/models/**/*.rb`)
-- Multiple patterns: comma-separated
-- Files without `applyTo` are ignored
-- Takes precedence over `copilot-instructions.md` when matched
-
-**excludeAgent Property:**
-- `excludeAgent: "code-review"` → Hidden from code review
-- `excludeAgent: "coding-agent"` → Hidden from coding agent
-- Omit property = both agents can use it
-
-**Best Practices:**
-- Scope guidance to individual areas (frontend, security, backend)
-- Use separate files per language/concern
-- Apply only coding agent to deployment/infrastructure code
-- Apply only code-review to PR-specific checks
-
-## 3. *.prompt.md Files
-
-**Location:** `.github/prompts/` folder
-**Scope:** Reusable task-specific templates, invoked via `/<filename>` slash command
-**Format:**
-```yaml
----
-mode: 'agent'
-description: 'Generate a new React form component'
-model: 'GPT-4o'  # Optional
-tools: ['githubRepo', 'search/codebase']  # Optional
----
-# Your prompt instructions
-```
-
-**Variables:**
-- `${input:varName:Placeholder text}` - User input fields
-- `{selectedText}` - Currently selected code in editor
-- Supports dynamic context injection
-
-**Key Features:**
-- Executable as slash commands (`/react-form-component`)
-- Available in VS Code, Visual Studio, JetBrains
-- Model specification allows GPT-4o override
-- Tool specification enables search, repo access
-
-**Best Practices:**
-- Create task-specific templates (code gen, testing, docs)
-- Use clear description for discoverability
-- Leverage `{selectedText}` for context-aware prompts
-- Reference via `#prompt:` in chat
-
-## 4. AGENTS.md
-
-**Location:** Any directory level (root recommended, nested supported)
-**Scope:** Directory tree downward; nested files override parent
-**Format:** Plain Markdown (any heading structure), no frontmatter required
-
-**Precedence Rules (Highest to Lowest):**
-1. Explicit user chat prompt
-2. Closest/most-deeply-nested AGENTS.md
-3. Root AGENTS.md
-4. `.github/copilot-instructions.md`
-
-**Nesting Behavior:**
-- Agents read nearest AGENTS.md in directory tree
-- Monorepos: Place nested AGENTS.md in package folders
-- Real example: OpenAI repo has 88 AGENTS.md files
-- Scope = entire directory tree rooted at file location
-
-**Implementation Notes:**
-- Plain text, no structured format required
-- Works alongside copilot-instructions.md & .instructions.md
-- Compatible with CLAUDE.md & GEMINI.md
-- **Note:** Known VS Code bug (Issue #271489) - only root-level currently respected in some versions; nested support inconsistent
-
-**Standardization:**
-- Now stewarded by Agentic AI Foundation (Linux Foundation)
-- Formalized August 2025
-- Spreading rapidly across OSS projects
-
-## 5. File Structure Hierarchy
-
-```
-.github/
-├── copilot-instructions.md          # Repo-wide base instructions
-├── instructions/
-│   ├── api.instructions.md          # applyTo: "src/api/**/*.ts"
-│   ├── tests.instructions.md        # applyTo: "**/*.test.ts"
-│   └── security.instructions.md     # excludeAgent: "code-review"
-└── prompts/
-    ├── generate-component.prompt.md
-    ├── write-tests.prompt.md
-    └── document-api.prompt.md
-
-AGENTS.md                            # Root-level, applies to all
-packages/
-├── web/
-│   └── AGENTS.md                    # Nested, overrides root for /web
-├── api/
-    └── AGENTS.md                    # Nested, overrides root for /api
-```
-
-## 6. Key Innovations (2025)
-
-**Path-Scoped Instructions (Sep 2025):**
-- `applyTo` glob patterns enable granular control
-- Reduces need for monolithic instruction files
-- Automatic application based on file context
-
-**Agent-Specific Instructions (Nov 2025):**
-- `excludeAgent` property allows per-agent configuration
-- Different rules for coding-agent vs code-review
-
-**Coding Agent AGENTS.md Support (Aug 2025):**
-- Previously chat-only, now applies to all agents
-- Formal standardization under Linux Foundation
-
-## 7. Organization-Level Configuration
-
-**GitHub Admin Settings:**
-- Enable/disable Copilot org-wide
-- Public code matching policies
-- Seat management & license assignment
-
-**VS Code Workspace Settings (.vscode/settings.json):**
-```json
-{
-  "github.copilot.enable": {
-    "*": true,
-    "markdown": false
-  },
-  "github.copilot.chat.codeGeneration.useInstructionFiles": true,
-  "chat.promptFilesLocations": [
-    ".github/prompts",
-    ".prompts"
-  ]
-}
-```
-
-## 8. Current Limitations & Compatibility
-
-**IDE Support:**
-- Prompt files: VS Code, Visual Studio, JetBrains only
-- Instructions: All IDEs via GitHub.com
-- Code review: GitHub.com only
-
-**Known Issues:**
-- Nested AGENTS.md parsing inconsistent (VS Code issue #271489)
-- Organization-level instruction rollout incomplete
-- Custom agent precedence edge cases documented
-
-## 9. Unresolved Questions
-
-1. Does `applyTo` support negative patterns (exclude) or only positive matching?
-2. Performance impact of 80+ nested AGENTS.md files in large monorepos?
-3. Interaction order when multiple `*.instructions.md` match same file?
-4. Can `excludeAgent` be used in `copilot-instructions.md` or only `*.instructions.md`?
-5. How does `tools` field in prompt frontmatter interact with available MCP servers?
-
-## Sources
-
-- [GitHub Copilot agent-specific instructions](https://github.blog/changelog/2025-11-12-copilot-code-review-and-coding-agent-now-support-agent-specific-instructions/)
-- [Path-scoped custom instruction support](https://github.blog/changelog/2025-09-03-copilot-code-review-path-scoped-custom-instruction-file-support/)
-- [Unlocking power of Copilot code review instructions](https://github.blog/ai-and-ml/unlocking-the-full-power-of-copilot-code-review-master-your-instructions-files/)
-- [VS Code custom instructions](https://code.visualstudio.com/docs/copilot/customization/custom-instructions)
-- [GitHub Copilot coding agent AGENTS.md support](https://github.blog/changelog/2025-08-28-copilot-coding-agent-now-supports-agents-md-custom-instructions/)
-- [How to write great agents.md](https://github.blog/ai-and-ml/github-copilot/how-to-write-a-great-agents-md-lessons-from-over-2500-repositories/)
-- [GitHub official prompt files tutorial](https://docs.github.com/en/copilot/tutorials/customization-library/prompt-files/your-first-prompt-file)
-- [Custom agents configuration](https://docs.github.com/en/copilot/reference/custom-agents-configuration)

package/plans/260108-0924-rebuild-cokit-awesome-copilot-format/phase-01-restructure-folders.md

@@ -1,183 +0,0 @@
-# Phase 1: Restructure Folders
-
-## Context Links
-- [Main Plan](./plan.md)
-- Reference: `C:\w\_demo\awesome-copilot-main\` (Awesome Copilot structure)
-
-## Overview
-
-| Priority | Status | Effort |
-|----------|--------|--------|
-| P1 | pending | 1h |
-
-## Key Insights
-
-1. Awesome Copilot dung cau truc flat voi 5 directories chinh
-2. Moi category co README rieng trong `docs/`
-3. Build scripts nam trong `eng/` folder
-4. Schema validation cho collections
-
-## Requirements
-
-### Target Directory Structure
-
-```
-C:\w\_me\cokit\
-├── agents/                  # *.agent.md files
-├── prompts/                 # *.prompt.md files
-├── instructions/            # *.instructions.md files
-├── skills/                  # */SKILL.md folders
-├── collections/             # *.collection.yml files
-├── docs/                    # Category READMEs
-│   ├── README.agents.md
-│   ├── README.prompts.md
-│   ├── README.instructions.md
-│   ├── README.skills.md
-│   └── README.collections.md
-├── eng/                     # Build scripts
-│   ├── constants.mjs
-│   ├── update-readme.mjs
-│   ├── validate-collections.mjs
-│   ├── validate-skills.mjs
-│   ├── create-collection.mjs
-│   ├── create-skill.mjs
-│   └── yaml-parser.mjs
-├── .schemas/                # JSON schemas
-│   └── collection.schema.json
-├── templates/               # Keep for init command (repo templates)
-│   └── repo/
-│       ├── .github/
-│       └── .vscode/
-├── src/                     # CLI source (keep existing)
-├── bin/                     # CLI entrypoint (keep existing)
-├── AGENTS.md                # Project instructions for AI
-├── README.md                # Main readme
-└── package.json
-```
-
-## Related Code Files
-
-| File | Purpose | Action |
-|------|---------|--------|
-| `package.json` | Dependencies | Add js-yaml, vfile-matter |
-| `.gitignore` | Git ignore | Update for new structure |
-
-## Implementation Steps
-
-### Step 1: Create Core Directories
-
-```bash
-# From C:\w\_me\cokit\
-mkdir agents
-mkdir prompts
-mkdir instructions
-mkdir collections
-mkdir eng
-mkdir .schemas
-```
-
-### Step 2: Create Collection Schema
-
-Create `.schemas/collection.schema.json`:
-
-```json
-{
-  "$schema": "http://json-schema.org/draft-07/schema#",
-  "title": "Collection Manifest",
-  "type": "object",
-  "required": ["id", "name", "description", "items"],
-  "properties": {
-    "id": {
-      "type": "string",
-      "pattern": "^[a-z0-9-]+$",
-      "maxLength": 50
-    },
-    "name": {
-      "type": "string",
-      "maxLength": 100
-    },
-    "description": {
-      "type": "string",
-      "maxLength": 500
-    },
-    "tags": {
-      "type": "array",
-      "items": { "type": "string", "pattern": "^[a-z0-9-]+$" },
-      "maxItems": 10
-    },
-    "items": {
-      "type": "array",
-      "minItems": 1,
-      "maxItems": 50,
-      "items": {
-        "type": "object",
-        "required": ["path", "kind"],
-        "properties": {
-          "path": { "type": "string" },
-          "kind": { "enum": ["prompt", "instruction", "agent", "skill"] }
-        }
-      }
-    }
-  }
-}
-```
-
-### Step 3: Update package.json
-
-Add dependencies:
-
-```json
-{
-  "dependencies": {
-    "commander": "^12.0.0",
-    "prompts": "^2.4.2",
-    "chalk": "^5.3.0",
-    "js-yaml": "^4.1.0",
-    "vfile-matter": "^5.0.0"
-  },
-  "scripts": {
-    "build": "node eng/update-readme.mjs",
-    "collection:validate": "node eng/validate-collections.mjs",
-    "collection:create": "node eng/create-collection.mjs",
-    "skill:validate": "node eng/validate-skills.mjs",
-    "skill:create": "node eng/create-skill.mjs"
-  }
-}
-```
-
-### Step 4: Update .gitignore
-
-```gitignore
-node_modules/
-*.log
-.DS_Store
-# Keep templates and build output
-!templates/
-!docs/README.*.md
-```
-
-### Step 5: Create Placeholder Files
-
-Create empty placeholder files de Git track directories:
-
-```bash
-touch agents/.gitkeep
-touch prompts/.gitkeep
-touch instructions/.gitkeep
-touch collections/.gitkeep
-```
-
-## Success Criteria
-
-- [ ] All 6 directories created
-- [ ] `.schemas/collection.schema.json` valid JSON
-- [ ] `package.json` updated with new deps and scripts
-- [ ] `.gitignore` updated
-- [ ] Directory structure matches Awesome Copilot format
-
-## Risk Assessment
-
-| Risk | Probability | Impact | Mitigation |
-|------|-------------|--------|------------|
-| Git conflicts | Low | Low | Create fresh directories |
-| Missing dependencies | Low | Medium | Test npm install |