skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
reason-machines/hermes-skills199 installs

hermes-agent-mission-control

Mission control Kanban interface for managing autonomous Hermes Agent tasks, routines, and workflows

How do I install this agent skill?

npx skills add https://github.com/reason-machines/hermes-skills --skill hermes-agent-mission-control
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    The skill provides documentation and setup instructions for 'Minions', a visual Kanban interface for managing Hermes Agent tasks. It involves installing a vendor-associated Node.js package and managing local SQLite databases. The identified features, such as task persistence and process management commands, are standard for the tool's stated purpose as a mission control interface.

  • Socketpass

    No alerts

  • Snykpass

    Risk: LOW · No issues

What does this agent skill do?

Hermes Agent Mission Control

Skill by 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

Quick start with npx (no installation required):

npx minionsai

Or install globally:

npm install -g minionsai

Then run:

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

# 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:

// 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
// 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

# 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

// 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

# 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:

// 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

# Default port is 6969
# If occupied, check for existing Minions process:
lsof -i :6969
kill -9 <PID>

# Then restart
npx minionsai

Database Issues

# 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:

# 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.

Limitations

  • Currently Hermes-only (OpenClaw adapter planned)
  • Notifications not yet implemented (roadmap)
  • Skills library not yet available (roadmap)

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/reason-machines/hermes-skills/hermes-agent-mission-control">View hermes-agent-mission-control on skillZs</a>