Explicit AI Assistant

A VS Code extension for AI-driven agile development with explicit context control. No auto-indexing, no hidden magic — you control exactly what the AI sees.

Features

Dual Workflow Modes

• Vibe Mode — Chat first, then build. Explore ideas and iterate as you discover needs.
• Agile Mode — Plan enough to start, then build, test, and improve continuously in small cycles.

Switch modes anytime from the chat UI. Conversations auto-save as threads when you switch.

AI-Powered Agile Planning

Full sprint-based agile workflow built into the chat:

Plans & Tasks

• Create plans with goals and break them into tasks
• Tasks support priority (high/medium/low), story points, acceptance criteria, and dependencies
• Auto-plan: describe a requirement and the AI suggests a task breakdown
• Dependency tracking: see which tasks are blocked vs ready to start

Sprints

• Create time-boxed sprints (default 14 days)
• Assign tasks to sprints
• Track velocity (story points completed per sprint)
• Sprint context automatically injected into AI prompts

Retrospectives

• Record what went well, what needs improvement, and action items
• Tied to sprints for historical tracking
• Inspect-and-adapt feedback loop

Agents

Custom AI personas stored as JSON in .explicitai/agents/:

• Create agents with custom system prompts (e.g., "Code Reviewer", "Architect", "Test Writer")
• Switch agents on the fly — the system prompt changes accordingly
• Agents persist across sessions

Built-in agents (always available):

Usage: type @ in the chat input to see available agents, then write your message:

@planner Build a user authentication system with OAuth2 and email/password
@reviewer #src/auth/login.ts check this for security issues
@retro we just finished the sprint, let's reflect

@ and # Syntax

• @agentName message — invoke a specific agent for this message (autocomplete on @)
• #path/to/file — attach a file to the conversation (autocomplete on #)
• @filename.ts — also attaches files (existing behavior)

Examples:

@planner I need a payment integration with Stripe
@reviewer #src/api/payments.ts #src/models/Order.ts review these
@tester #src/auth/login.ts what tests do I need?
How does #src/core/planner.ts work?

Project Memory

The AI remembers across sessions:

• Decisions, preferences, patterns, warnings, and knowledge
• Keyword-based retrieval — relevant memories auto-injected into prompts
• Manual remember/forget from the chat UI

Built-in Developer Tools

The AI can use workspace tools directly:

Read tools execute immediately. Write/shell tools always require your approval.

MCP Integration

Model Context Protocol support with manual approval for every action:

• Configure servers in .explicitai/mcp.json
• Toggle individual tools (filesystem, terminal, HTTP)
• Enable/disable servers from the UI

Commands

Thread History

• Conversations auto-save as threads
• Switch between threads or revert to a previous state
• Threads track mode (vibe/agile) and active agent

Context Management

• Token budget tracking with visual utilization bar
• Automatic conversation compaction (summarizes old messages to free space)
• Manual compact button when you want to free context immediately
• Stack detection auto-injects project info (framework, language, test runner, etc.)

Setup

1. Install LM Studio and load a model
2. Start the local server (default: http://localhost:1234)
3. Install this extension
4. Open the Explicit AI sidebar from the activity bar

Configuration

All settings under explicitAI.* in VS Code settings:

Agile Workflow Guide

Quick Start (Agent-Driven)

The fastest way to do agile development:

@planner Build a REST API for user management with CRUD operations

The planner agent breaks it into tasks with story points and acceptance criteria. Then:

@sprint-master create a sprint and pull in the high-priority tasks
@sprint-master what should I work on next?
@reviewer #src/controllers/UserController.ts
@retro we finished the sprint

Manual Workflow

1. Switch to Agile mode in the chat
2. Create a plan: give it a title and goal
3. Use @planner with your requirement — the AI breaks it into tasks
4. Create a sprint and assign tasks
5. Start the sprint
6. Work through tasks, updating status as you go
7. Complete the sprint — check velocity
8. Run @retro for a retrospective

Message API (for webview integration)

Plans:

{ "type": "createPlan", "title": "Auth System", "goal": "Secure user authentication" }
{ "type": "getPlan" }

Tasks:

{ "type": "addDetailedTask", "title": "Login endpoint", "description": "POST /auth/login", "priority": "high", "storyPoints": 5, "acceptanceCriteria": ["Returns JWT", "Rate limited"], "dependsOn": [] }
{ "type": "updateTaskStatus", "taskId": "task_xxx", "status": "done" }
{ "type": "setStoryPoints", "taskId": "task_xxx", "points": 3 }
{ "type": "setAcceptanceCriteria", "taskId": "task_xxx", "criteria": ["Passes tests"] }
{ "type": "setDependencies", "taskId": "task_xxx", "dependsOn": ["task_yyy"] }

Sprints:

{ "type": "createSprint", "name": "Sprint 1", "goal": "Core auth", "durationDays": 7 }
{ "type": "addTaskToSprint", "sprintId": "sprint_xxx", "taskId": "task_xxx" }
{ "type": "startSprint", "sprintId": "sprint_xxx" }
{ "type": "completeSprint", "sprintId": "sprint_xxx" }

Retrospectives:

{ "type": "addRetrospective", "wentWell": ["Fast delivery"], "needsImprovement": ["Test coverage"], "actionItems": ["Add CI checks"], "sprintId": "sprint_xxx" }
{ "type": "getRetrospectives" }

Auto-planning:

{ "type": "autoPlan", "requirement": "Build OAuth2 login with Google and GitHub providers" }

Architecture

src/
├── agents/          # Custom AI agent registry
├── chat/            # Chat session management
├── commands/        # VS Code command implementations
├── core/            # Core modules (planner, memory, context, LLM client, etc.)
├── mcp/             # Model Context Protocol integration
├── threads/         # Conversation thread persistence
├── tools/           # Built-in workspace tools
├── ui/              # Webview host, panel, provider
└── extension.ts     # Entry point

Data stored in workspace:

.explicitai/
├── agents/          # Agent JSON configs
├── threads/         # Conversation snapshots
├── plans.json       # Plans, tasks, sprints, retros
└── memory.json      # Persistent project memory

Development

npm install
npm run build        # Compile TypeScript + validate HTML
npm run watch        # Watch mode for development

Press F5 in VS Code to launch the extension in a development host.

License

Local use only.