- name
- hermes-agent-mission-control
- description
- Mission control Kanban interface for managing autonomous Hermes Agent tasks, routines, and workflows
- triggers
- ["set up minions mission control for hermes agent","create a kanban board for hermes tasks","manage multiple hermes agent sessions","track autonomous agent work with minions","install hermes mission control interface","organize hermes agent tasks in a board","run minions for hermes agent management","set up task supervision for hermes"]
# Hermes Agent Mission Control
> Skill by [ara.so](https://ara.so) — Hermes Skills collection
Minions is a mission control interface for Hermes Agent that provides a Kanban board to create, supervise, and review autonomous work. It solves the problem of juggling multiple terminal sessions and losing track of long-running tasks by providing a visual board with task status (in progress, in review, done), live streaming of agent work, and human-in-the-loop verification.
## Installation
**Prerequisites:** Node.js 18+ and [Hermes Agent](https://hermes-agent.nousresearch.com)
Quick start with npx (no installation required):
```bash
npx minionsai
```
Or install globally:
```bash
npm install -g minionsai
```
Then run:
```bash
minions
```
The server starts on `http://localhost:6969` by default.
## Data Storage
Minions creates a local SQLite database on first run:
- Database and state stored in `~/.minions/`
- Chat transcripts live in Hermes's session database
- Task metadata, status, and settings stored in Minions SQLite DB
- Completely local-first by default (no cloud dependency)
## CLI Commands
```bash
# Start the server
minions
# Check version
minions --version
# Compare with latest NPM version
npm view minionsai version
```
## Key Features & Usage
### Task Management
Each task in Minions is a persistent Hermes root session:
1. **Create tasks**: Describe what you want in chat
2. **Autonomous execution**: Agent decides how to get it done
3. **Live streaming**: Watch tool calls, reasoning, and responses in real-time
4. **Completion judge**: Lightweight LLM evaluates whether task is done after each turn
5. **Human verification**: Review agent's proposed completion before moving to "Done"
### Kanban Board States
- **In Progress**: Tasks actively being worked on
- **In Review**: Tasks completed by agent, awaiting human verification
- **Done**: Verified and closed tasks
### Per-Task Configuration
Override model and reasoning settings on individual tasks:
```typescript
// Task metadata structure
interface Task {
id: string;
title: string;
description: string;
status: 'in_progress' | 'in_review' | 'done';
sessionId: string; // Hermes session ID
model?: string; // Override default model
reasoningEffort?: number; // Override reasoning effort
createdAt: string;
updatedAt: string;
}
```
### Routines
Create and manage recurring Hermes jobs:
- Schedule regular agent tasks
- Track execution history
- Review output from each run
- Useful for monitoring, data collection, reporting
### File Browser
View files created by agents in the workspace directory without leaving the interface.
## Configuration
Check settings in the web UI Settings page:
- View running Minions server version
- Configure Hermes Agent connection
- Set default model and reasoning effort
- Manage workspace directory path
## Architecture
Minions is built with TypeScript and uses:
- SQLite for local task metadata storage
- Hermes Agent sessions for chat transcripts
- Real-time WebSocket connections for live streaming
- Kanban board UI for task visualization
```typescript
// Example: Task state transitions
// 1. User creates task → status: 'in_progress'
// 2. Agent completes work → status: 'in_review'
// 3. Human verifies → status: 'done'
// The completion judge runs after each agent turn
interface CompletionCheck {
isComplete: boolean;
reasoning: string;
confidence: number;
}
```
## Common Patterns
### Delegating Long-Running Work
```bash
# Start Minions
npx minionsai
# In the web UI:
# 1. Create new task: "Research top 10 AI tools launched this month"
# 2. Walk away - agent executes autonomously
# 3. Return later to review in "In Review" column
# 4. Verify and move to "Done"
```
### Managing Multiple Projects
```typescript
// Each task is independent:
// - Separate Hermes session
// - Independent model/effort settings
// - Parallel execution supported
// Example workflow:
// Task 1: "Write blog post about TypeScript 5.4 features"
// Task 2: "Analyze competitor pricing pages"
// Task 3: "Generate test data for user onboarding flow"
// All run concurrently, visible on one board
```
### Routine Monitoring
```bash
# Set up daily routine:
# "Check GitHub stars and create summary report"
# Minions will:
# 1. Execute on schedule
# 2. Store execution history
# 3. Make output reviewable
# 4. Alert if intervention needed (roadmap feature)
```
## Integration with Hermes Agent
Minions acts as a control layer over Hermes:
```typescript
// Minions doesn't replace Hermes CLI usage
// It complements it by providing:
// - Visual task organization
// - Persistent session management
// - Completion detection
// - Human review workflow
// You can still use Hermes directly:
hermes chat "quick question"
// Or use Minions for longer work:
// Create task in UI → Minions manages Hermes session
```
## Troubleshooting
### Port Already in Use
```bash
# Default port is 6969
# If occupied, check for existing Minions process:
lsof -i :6969
kill -9 <PID>
# Then restart
npx minionsai
```
### Database Issues
```bash
# Database location
ls -la ~/.minions/
# Reset database (WARNING: deletes all tasks)
rm -rf ~/.minions/
npx minionsai # Creates fresh database
```
### Connection to Hermes Agent
Ensure Hermes Agent is properly installed and accessible:
```bash
# Verify Hermes installation
which hermes
hermes --version
# Check Hermes can start sessions
hermes chat "test"
```
### Task Stuck in "In Progress"
- Check Hermes session logs for errors
- Review agent's last output in live stream
- Manually intervene in chat
- Adjust model/reasoning effort settings
- Close and recreate task if necessary
## Hosted Option
Self-hosting is recommended for privacy, but a hosted option is available at [Agent37](https://www.agent37.com).
## Limitations
- Currently Hermes-only (OpenClaw adapter planned)
- Notifications not yet implemented (roadmap)
- Skills library not yet available (roadmap)
GitHubで見る