Mysti Features Guide

March 11, 2026 · View on GitHub

Detailed documentation for all Mysti features.

Chat Interface
Providers
Brainstorm Mode
Autonomous Mode
@-Mention System
Context Compaction
Agent Configuration
Plan Selection
Permission System
Agent Lifecycle
Slash Commands
Settings Reference

Chat Interface

Basic Usage

The Mysti chat interface is accessible from the Activity Bar sidebar or as a standalone editor tab.

Opening the Chat:

Click the Mysti icon in the Activity Bar
Use keyboard shortcut: Ctrl+Shift+M (Windows/Linux) or Cmd+Shift+M (Mac)
Run command: Mysti: Open Mysti Chat

Opening in New Tab:

Use keyboard shortcut: Ctrl+Shift+N (Windows/Linux) or Cmd+Shift+N (Mac)
Click the "Open in New Tab" button in the chat header

Context Management

Add relevant code context to improve AI responses.

Adding Context:

Right-click a file in Explorer -> "Add to Mysti Context"
Right-click selected code in Editor -> "Add to Mysti Context"
Use @filename mentions for transient file context

Auto-Context: When enabled (mysti.autoContext: true), Mysti automatically tracks your active editor and includes relevant context.

Per-Panel Isolation: Each chat panel (sidebar or tab) maintains its own independent context. Adding a file to one panel doesn't affect others.

Conversation Management

New Conversation:

Click the "New" button in the header
Run command: Mysti: New Conversation

Conversation History:

Access previous conversations from the history panel
Each conversation persists across VSCode sessions
Per-conversation settings (mode, model, provider) are preserved

Providers

Mysti supports 12 AI providers through their CLI interfaces. See PROVIDERS.md for complete setup guides.

Provider	Best For
Claude Code	Deep reasoning, complex refactoring
Codex	Quick iterations, OpenAI ecosystem
Gemini	Fast responses, Google integration
GitHub Copilot	Multi-model access via GitHub subscription
Cline	Plan/Act workflow, structured tasks
Cursor	Auto model selection, multi-model
OpenClaw	WebSocket streaming, thinking levels
Manus (experimental)	HTTP API-based, async tasks

Switching Providers:

Use the /agent slash command
Change mysti.defaultProvider in settings
Each conversation remembers its provider

Brainstorm Mode

Brainstorm mode enables two AI agents to collaborate using structured reasoning frameworks. See BRAINSTORM.md for the full guide.

Collaboration Strategies

Strategy	Roles	Best For
Quick	Direct synthesis	Simple tasks
Debate	Critic vs Defender	Architecture decisions
Red-Team	Proposer vs Challenger	Security reviews
Perspectives	Risk Analyst vs Innovator	Greenfield design
Delphi	Facilitator vs Refiner	Complex consensus

Convergence Detection

During discussion rounds, Mysti tracks agent agreement and position stability. When auto-convergence is enabled, discussions exit early once agents reach consensus.

Configuration

{
  "mysti.brainstorm.enabled": true,
  "mysti.brainstorm.strategy": "debate",
  "mysti.brainstorm.autoConverge": true,
  "mysti.brainstorm.maxDiscussionRounds": 3
}

Autonomous Mode

Let the AI work independently with safety controls. See AUTONOMOUS-MODE.md for the full guide.

Safety Classification

Every operation is classified before auto-approval:

Level	Action	Examples
Safe	Auto-approve	File reads, test commands, git status
Caution	Mode-dependent	File creates/edits, unknown bash commands
Blocked	Always deny	File deletion, force push, sudo

Safety Modes

Conservative: Only clearly safe operations auto-approved
Balanced (default): Common development operations auto-approved
Aggressive: Everything auto-approved except hardcoded blocks

Memory System

Mysti learns from your decisions:

Permission approvals/denials are remembered
Question answers are cached for auto-response
Confidence decays over time (0.95/day)
Old memories auto-pruned when capacity reached

Continuation Modes

Goal: Free-form goal with completion detection
Task Queue: Sequential task list with progress tracking

@-Mention System

Route tasks to specific agents and reference files inline. See MENTIONS.md for the full guide.

File Mentions

@utils.ts Can you explain these helper functions?

Adds the file as transient context (not persisted).

Agent Mentions

@claude Review this code for security issues
@claude Write tests, then @gemini review them

Routes tasks to specific agents. Later agents receive earlier agents' responses as context.

Switching Providers

Switch to @cursor

Context Compaction

Automatic context management to prevent overflow. See COMPACTION.md for the full guide.

Tracks per-panel token usage
Triggers at configurable threshold (default 75%)
Native CLI strategy for Claude Code (/compact)
Client-side summarization for other providers

Agent Configuration

Customize AI behavior with personas and skills. See PERSONAS-AND-SKILLS.md for the full guide.

16 built-in personas: Architect, Prototyper, Product-Centric, Refactorer, DevOps, Domain Expert, Researcher, Builder, Debugger, Integrator, Mentor, Designer, Full-Stack, Security-Minded, Performance Tuner, Toolsmith.

Toggleable Skills

12 skills: Concise, Repo Hygiene, Organized, Auto-Commit, First Principles, Auto-Compact, Dependency-Aware, Graceful Degradation, Scope Discipline, Doc Reflexes, Test-Driven, Rollback Ready.

Three-Tier Loading

Agent definitions load progressively for fast UI:

Tier 1: Metadata (always loaded for lists)
Tier 2: Instructions (loaded on selection)
Tier 3: Full content (loaded on demand)

Custom Agents

Create custom personas and skills as markdown files in:

.mysti/agents/ (workspace, highest priority)
~/.mysti/agents/ (user)
resources/agents/core/ (bundled, lowest priority)

Plan Selection

When the AI presents multiple implementation options, Mysti detects and displays them for selection.

How Plan Detection Works

AI response contains implementation options
ResponseClassifier analyzes the response
Detected plans displayed as selectable cards
User selects preferred approach and execution mode

Plan Card Information

Each plan card shows:

Title - Name of the approach
Summary - Brief description
Pros - Advantages
Cons - Trade-offs
Complexity - Low, Medium, or High

Execution Modes

Mode	Description
Ask Before Edit	AI explains changes and waits for approval
Edit Automatically	AI makes changes directly
Plan	AI creates detailed plan without making changes

Permission System

Control what actions the AI can perform.

Access Levels

Level	Description
Read-only	AI cannot modify any files
Ask-permission	AI requests approval for each action
Full-access	AI can perform all operations

Permission Requests

In ask-permission mode, the AI shows requests for:

File creation, editing, deletion
Bash commands
Web requests

Each request includes action description, risk level, diff preview, and approve/deny buttons.

Timeout Behavior

Behavior	Description
`auto-accept`	Automatically approve after timeout
`auto-reject`	Automatically deny after timeout
`require-action`	No timeout, require explicit action

{
  "mysti.lifecycle.enabled": true,
  "mysti.lifecycle.idleTimeoutMinutes": 60,
  "mysti.lifecycle.processTreeTracking": true,
  "mysti.lifecycle.protectActiveChildren": true
}

Slash Commands

Type / in the chat to access slash commands organized by section.

Sections

Section	Commands
Context	`/context`, `/clear`
Model	`/model`, `/agent`, `/mode`
Customize	`/persona`, `/skills`
Commands	`/compact` (Claude), `/thinking` (Claude), `/profile` (Codex), `/plan-act` (Cline)
Settings	`/access`, `/thinking-level`
Support	`/help`

Settings Reference

Provider Settings

Setting	Default	Description
`mysti.defaultProvider`	`"claude-code"`	Default AI provider
`mysti.claudePath`	`"claude"`	Path to Claude CLI
`mysti.codexPath`	`"codex"`	Path to Codex CLI
`mysti.geminiPath`	`"gemini"`	Path to Gemini CLI
`mysti.copilotPath`	`"copilot"`	Path to Copilot CLI
`mysti.clinePath`	`"cline"`	Path to Cline CLI
`mysti.cursorPath`	`"agent"`	Path to Cursor CLI
`mysti.openclawPath`	`"openclaw"`	Path to OpenClaw CLI

Operation Settings

Setting	Default	Description
`mysti.defaultMode`	`"ask-before-edit"`	Default operation mode
`mysti.defaultThinkingLevel`	`"medium"`	Thinking level (none/low/medium/high)
`mysti.accessLevel`	`"ask-permission"`	File operation access level
`mysti.autoContext`	`true`	Auto-include relevant context

Brainstorm Settings

Setting	Default	Description
`mysti.brainstorm.enabled`	`false`	Enable brainstorm mode
`mysti.brainstorm.agents`	`["claude-code", "openai-codex"]`	Which 2 agents to use
`mysti.brainstorm.strategy`	`"quick"`	Collaboration strategy
`mysti.brainstorm.autoConverge`	`true`	Auto-exit when agents converge
`mysti.brainstorm.maxDiscussionRounds`	`3`	Maximum discussion rounds

Autonomous Settings

Setting	Default	Description
`mysti.autonomous.safetyMode`	`"balanced"`	Safety mode
`mysti.autonomous.maxSessionDuration`	`24`	Max session hours
`mysti.autonomous.allowFileCreation`	`true`	Allow file creation
`mysti.autonomous.allowFileEdit`	`true`	Allow file editing
`mysti.autonomous.allowBashCommands`	`true`	Allow bash commands
`mysti.autonomous.blockPatterns`	`[]`	Custom block patterns

Compaction Settings

Setting	Default	Description
`mysti.compaction.enabled`	`true`	Enable context compaction
`mysti.compaction.threshold`	`75`	Compaction threshold (%)

Lifecycle Settings

Setting	Default	Description
`mysti.lifecycle.enabled`	`true`	Enable lifecycle management
`mysti.lifecycle.idleTimeoutMinutes`	`60`	Idle timeout
`mysti.lifecycle.processTreeTracking`	`true`	Track child processes
`mysti.lifecycle.protectActiveChildren`	`true`	Protect active children

Agent Settings

Setting	Default	Description
`mysti.agents.autoSuggest`	`true`	Auto-suggest personas
`mysti.agents.maxTokenBudget`	`0`	Max tokens (0 = unlimited)

Permission Settings

Setting	Default	Description
`mysti.permission.timeout`	`30`	Timeout seconds (0 = none)
`mysti.permission.timeoutBehavior`	`"auto-reject"`	Timeout behavior