AI Skeleton Prompts & Memory

⚠️ WORK IN PROGRESS: This extension is under active development. Features may not work fully, may break, or may be removed/changed without notice. Use at your own risk.

Complete AI agent toolkit for VS Code: Native memory management, workflow prompts, MCP integration, and agent definitions.

🎯 Purpose

This extension extends Copilot agent context and memory capabilities by providing:

AI-Memory Tools - 7 Language Model Tools that agents can invoke autonomously
Workflow Prompts - Structured prompts for Think → Plan → Execute workflows
MCP Integration - Easy installation of Model Context Protocol servers
Agent Definitions - Pre-configured agent modes (Memory & Deep Thinking)

📦 Installation

Recent updates

appendEntry() ID retrieval (sql.js): fixed a backend edge-case where append operations could return null. The implementation now uses a dual-strategy lookup (primary: last_insert_rowid(), fallback: MAX(id)) to reliably return saved entry IDs.
ShowMemoryTool now uses selectContextForBudget() to provide token-budgeted, semantically-relevant context (default token budget ≈ 50k) rather than dumping the entire DB into prompts.
scripts/sync-agent-tools.js now emits only fully-qualified marketplace tool IDs (e.g. JasdeepN.ai-skeleton-extension/showMemory) in embedded agent frontmatter. Runtime tool registration remains aiSkeleton_* for the extension.
New @aiSkeleton chat participant with guaranteed LM tool invocation and token tracking; improved Memory Dashboard UX with read-only entry viewer and phase tracking.

Requirements

VS Code: 1.95.0 or newer (for Language Model Tools API)
License: Proprietary — see LICENSE (explicit written approval required)

Install from VS Code Marketplace (Recommended)

Search "AI Skeleton" in VS Code Extensions
Click Install on "AI Skeleton Prompts & Memory" by JasdeepN

Or install via command line:

code --install-extension JasdeepN.ai-skeleton-extension

Install from VSIX

Download latest .vsix from releases
In VS Code: Extensions → ... menu → "Install from VSIX..."
Select the downloaded .vsix file
Reload VS Code

Verify Installation

Run command: AI Skeleton: List Registered LM Tools

You should see 7 registered tools including showMemory, logDecision, updateContext, etc.

🚀 Quick Start

1. Install Agent Assets

Run command:

AI Skeleton: Install All

This installs:

Prompts to .github/prompts/
Agents to .github/agents/
Protected files to .github/
MCP configuration to .vscode/mcp.json

Tip: After installing MCP config, you can auto-start MCP servers on startup. See Settings → AI Skeleton: MCP Auto Start.

2. Verify Memory Tools

In Copilot chat, type # and you should see the memory tools appear in the picker:

#showMemory - Read AI-Memory contents
#logDecision - Log decisions
#updateContext - Update active context
And 4 more...

3. Use Agent Mode

Switch to agent mode in Copilot and select the Memory & Deep Thinking agent from .github/agents/.

📦 Features

1. AI-Memory Tools (Language Model Tools)

Agents can autonomously invoke these tools on VS Code 1.95+:

Note: At runtime the extension registers tools under aiSkeleton_* names (e.g. aiSkeleton_showMemory). For release artifacts and embedded agent frontmatter, the sync/embedding process now writes fully-qualified marketplace tool IDs of the form JasdeepN.ai-skeleton-extension/<toolReferenceName> (for example JasdeepN.ai-skeleton-extension/showMemory). This keeps release assets compatible with Marketplace tooling and agent validation.

Tool	Purpose	Parameters
`aiSkeleton_showMemory`	Read memory bank contents	`fileType` (optional): "all", "activeContext", "decisionLog", etc.
`aiSkeleton_logDecision`	Log architectural/technical decisions	`decision`, `rationale`
`aiSkeleton_updateContext`	Update active working context	`contextUpdate`
`aiSkeleton_updateProgress`	Track task progress	`item`, `status` ("done", "doing", "next")
`aiSkeleton_updatePatterns`	Record system patterns/conventions	`pattern`, `context`
`aiSkeleton_updateProjectBrief`	Update project goals/scope	`briefUpdate`
`aiSkeleton_markDeprecated`	Mark items deprecated without deletion	`item`, `reason`, `replacement`

Example Agent Usage:

// Agent automatically invokes:
aiSkeleton_logDecision({
  decision: "Use TypeScript for type safety",
  rationale: "Catches errors at compile time"
})

aiSkeleton_updateProgress({
  item: "Implement user authentication",
  status: "done"
})

2. @aiSkeleton Chat Participant

New in v0.2.35: A dedicated chat participant that guarantees memory tool invocation and automatic token tracking.

How It Works:

Type @aiSkeleton in Copilot chat to interact with a memory-aware AI assistant:

@aiSkeleton What was our last database optimization decision?

The @aiSkeleton participant:

Automatically uses memory tools - Invokes showMemory, logDecision, etc. based on conversation context
Tracks token usage - Every tool invocation is logged for visibility
Provides suggestions - Followup suggestions for related memories and actions
Handles errors gracefully - Shows helpful messages if Copilot unavailable or quota exceeded

Example Conversation:

You: @aiSkeleton What patterns have we documented for database queries?

@aiSkeleton:
1. Invokes aiSkeleton_showMemory tool
2. Retrieves patterns tagged with "database"
3. Summarizes key patterns
4. Suggests: "Log a Decision about query optimization?"

Benefits:

✅ Natural language interface to project memory
✅ Automatic token counting per tool invocation
✅ Memory-aware responses from AI
✅ Guaranteed tool invocation (unlike default Copilot)
✅ Persistent context across conversations

Status Bar Indicators:

📊 Context Budget - Shows remaining token budget (healthy/warning/critical)
Click to see detailed token usage breakdown

3. Memory Bank Structure

AI-Memory/
├── activeContext.md      # Current work focus
├── decisionLog.md        # Architectural decisions
├── progress.md           # Task tracking (done/doing/next)
├── systemPatterns.md     # Code patterns/conventions
└── projectBrief.md       # Project goals/context

4. Workflow Prompts

Embedded (baked) versions of:

Think.prompt.md - Deep research and analysis
Plan.prompt.md - Task breakdown and planning
Execute.prompt.md - Implementation guidance
Checkpoint.prompt.md - Progress review
Sync.prompt.md - Memory-codebase synchronization
GH.prompt.md - GitHub Actions workflows
Startup.prompt.md - Session initialization

Optional workspace override: if your repository has .github/prompts/*.prompt.md files they will automatically be used (when aiSkeleton.prompts.source = auto or workspace).

5. Memory & Deep Thinking Agent

Pre-configured agent mode (memory-deep-think.agent.md) with:

Autonomous memory updates
Structured reasoning workflows
MCP tool integration
Tagged entry system for efficient scanning ([TYPE:YYYY-MM-DD] format)

6. Visual Interface

Memory Bank Tree View: Visual browser for AI-Memory/ files in Explorer
AI Skeleton Prompts Tree View: Browse and interact with embedded prompts
Memory Status Bar: Shows ACTIVE/INACTIVE status; click to create or view
Commands & Tools: Both manual commands and agent-invokable tools available

🛠️ Usage

Creating Memory Bank

Open Command Palette (Ctrl+Shift+P)
Run: AI Skeleton: Create Memory Bank
Select workspace folder
Memory bank files created in AI-Memory/

Installing Prompts & Agents

Command Palette → AI Skeleton: Install All

Installs to .github/prompts/ and .github/agents/ in your workspace.

Agent Memory Integration

On VS Code 1.95+, agents can autonomously use memory tools. Use Extension Development Host (F5) for development/debugging if needed:

# .github/agents/memory-deep-think.agent.md
tools:
  - JasdeepN.ai-skeleton-extension/showMemory
  - JasdeepN.ai-skeleton-extension/logDecision
  - JasdeepN.ai-skeleton-extension/updateContext
  - JasdeepN.ai-skeleton-extension/updateProgress
  - JasdeepN.ai-skeleton-extension/updatePatterns
  - JasdeepN.ai-skeleton-extension/updateBrief
  - JasdeepN.ai-skeleton-extension/markDeprecated

Agent automatically invokes tools during conversation. No manual intervention needed.

📋 Commands

Prompts

AI Skeleton: List Prompts - Quick pick prompt browser
AI Skeleton: Save Prompt to File... - Save to disk
AI Skeleton: Insert Prompt at Cursor - Insert into editor
AI Skeleton: Copy Prompt to Clipboard - Copy content
AI Skeleton: Install Built-in Prompts - Install all to .github/prompts
AI Skeleton: Install Single Prompt - Install one prompt

Installation

AI Skeleton: Install Agent Templates - Install to .github/agents
AI Skeleton: Install Protected Files - Install to .github/
AI Skeleton: Install All - One-click full setup
AI Skeleton: Install MCP Servers - Add .vscode/mcp.json with recommended servers

Memory Management (Manual)

AI Skeleton: Memory Bank Status - Check/create memory bank
AI Skeleton: Create Memory Bank - Initialize memory-bank folder
AI Skeleton: Show Memory Summary - View all memory contents
AI Skeleton: Log Decision - Record a decision
AI Skeleton: Update Context - Update active context
AI Skeleton: Update Progress - Track progress items
AI Skeleton: Update System Patterns - Record patterns
AI Skeleton: List Registered LM Tools - Verify agent tools availability

MCP

AI Skeleton: Start MCP Servers — Reads .vscode/mcp.json and starts configured command-based MCP servers in terminals

Settings

aiSkeleton.mcp.autoStart (boolean): Automatically start MCP servers on window startup. Prompts on first run per-workspace.
aiSkeleton.prompts.source: auto | embedded | workspace

Configuration

aiSkeleton.prompts.source: auto | embedded | workspace

🤔 FAQ

Why doesn't it work when I install the VSIX?

It should, provided you're on VS Code 1.95+. The Language Model Tools API is stable in these versions.

If tools don't appear in agent mode or the tool picker:

Verify VS Code version (Help → About): must be 1.95 or newer
Reload VS Code after installation
Run AI Skeleton: List Registered LM Tools to verify 7 tools are registered

MCP server fails with `ModuleNotFoundError: pydantic_core`

Some systems load extra Python site-packages (e.g., from SDK toolchains) that clash with uvx-installed MCP servers. Run MCP servers with a clean Python env:

scripts/uvx-clean.sh mcp-server-git --help
scripts/uvx-clean.sh mcp-server-git -r /path/to/repo
scripts/uvx-clean.sh mcp-server-fetch --help

The wrapper unsets PYTHONPATH and enables PYTHONNOUSERSITE=1, preventing conflicting site-packages from being injected.

Note: When you start MCP servers via the extension (using .vscode/mcp.json), uvx commands are automatically wrapped with this clean environment so you don't have to do anything extra inside VS Code.

Can I use this without the extension?

Yes! The prompts and agents work standalone:

Copy .github/prompts/ and .github/agents/ to your project
Agents can still read/write AI-Memory/ files via filesystem MCP
The extension adds enhanced UX and autonomous tool invocation

What about production use?

For production VSIX distribution, you would need:

Publish to VS Code Marketplace
Request proposed API access from Microsoft
Get publisher ID allowlisted

Or, use the file-based approach (agents read/write memory files directly via MCP filesystem tool).

How do I know if tools are available?

Run command: AI Skeleton: List Registered LM Tools

Expected: 7 tools registered (showMemory, logDecision, updateContext, updateProgress, updatePatterns, updateBrief, markDeprecated).

🔧 Development

Building

npm install
npm run build  # Embeds assets + compiles TypeScript

Press F5 in VS Code to launch an Extension Development Host and test the extension.

Packaging

npm run package:vsix
# or
node scripts/package-vsix.js

Automated versioning: auto-bumps patch version (0.0.1) if unchanged since last package.

Project Structure

ai-skeleton-extension/
├── src/
│   ├── extension.ts          # Main activation
│   ├── memoryTools.ts         # LM Tools registration (7 tools)
│   ├── memoryService.ts       # Core memory operations
│   ├── promptStore.ts         # Embedded prompts (auto-generated)
│   ├── agentStore.ts          # Embedded agents (auto-generated)
│   └── mcpStore.ts            # MCP configurations
├── agents/
│   └── memory-deep-think.agent.md  # Agent definition (source)
├── prompts/
│   ├── Think.prompt.md
│   ├── Plan.prompt.md
│   └── Execute.prompt.md
│   └── ...
├── scripts/
│   ├── embed-prompts.js       # Build-time embedding
│   ├── embed-agents.js        # Build-time embedding
│   └── package-vsix.js        # Auto-versioning packager
└── package.json               # Extension manifest

📖 Documentation

INSTALLATION.md - Setup guide and troubleshooting
LANGUAGE_MODEL_TOOLS_FIX.md - Technical details on LM Tools API
ARCHITECTURE_v0.1.3.md - Command-based architecture attempt (deprecated)
memory-bank/decisionLog.md - Full decision history

🧪 Testing

Quick CLI checks:

npm install
npm run build  # Full build with asset embedding
npm run verify # Verify embedded prompts

Manual interactive tests (in VS Code):

Open repository in VS Code (ai-skeleton-extension folder)
Press F5 to start Extension Development Host
In the new window:
- Check Status Bar for AI Prompts: N and Memory: ACTIVE/INACTIVE
- Run AI Skeleton: List Registered LM Tools → verify 7 tools
- Expand AI Skeleton Prompts in Explorer
- Create Memory Bank → Log Decision → Show Summary
- Test agent integration with Copilot chat

📝 Memory Management Rules (Enforced)

Never delete from memory files - only append new entries
Tag all entries with [TYPE:YYYY-MM-DD] format for efficient scanning
Timestamp everything automatically
Read selectively: Use tags/timestamps to scan only recent/relevant entries

📄 License

Proprietary. See LICENSE. Usage requires explicit written approval by the publisher.

🙏 Credits

Built for agent-assisted development workflows. Uses VS Code's Language Model Tools API for seamless agent integration.

Marketplace: JasdeepN.ai-skeleton-extension
Install: code --install-extension JasdeepN.ai-skeleton-extension
Status: Production-ready on VS Code 1.95+

AI Skeleton Prompts & Memory

JasdeepN

AI Skeleton Prompts & Memory

🎯 Purpose

📦 Installation

Recent updates

Requirements

Install from VS Code Marketplace (Recommended)

Install from VSIX

Verify Installation

🚀 Quick Start

1. Install Agent Assets

2. Verify Memory Tools

3. Use Agent Mode

📦 Features

1. AI-Memory Tools (Language Model Tools)

2. @aiSkeleton Chat Participant

3. Memory Bank Structure

4. Workflow Prompts

5. Memory & Deep Thinking Agent

6. Visual Interface

🛠️ Usage

Creating Memory Bank

Installing Prompts & Agents

Agent Memory Integration

📋 Commands

Prompts

Installation

Memory Management (Manual)

MCP

Settings

Configuration

🤔 FAQ

Why doesn't it work when I install the VSIX?

MCP server fails with ModuleNotFoundError: pydantic_core

Can I use this without the extension?

What about production use?

How do I know if tools are available?

🔧 Development

Building

Packaging

Project Structure

📖 Documentation

🧪 Testing

📝 Memory Management Rules (Enforced)

📄 License

🙏 Credits

MCP server fails with `ModuleNotFoundError: pydantic_core`