Skill: Dispatch

Owner: Neidraidd

Summary: AI-native project management with cost estimates, phased execution, multi-agent coordination, and privacy-aware model selection. Use for complex multi-step work requiring cost transparency and agent orchestration.

Tags: budget:1.0.0, cost-tracking:1.0.0, latest:1.0.0, phases:1.0.0, project-management:1.0.0

Version history:

v1.0.0 | 2026-02-08T12:01:13.052Z | user

Initial release of the Dispatch skill for AI-native project management in OpenClaw.

• Adds multi-phase, multi-agent project management with natural language commands.
• Provides upfront cost estimates and agent auto-assignment per project phase.
• Supports privacy-aware agent/model selection for sensitive data.
• Installation workflow requires explicit user confirmation before any changes.
• Includes detailed config file, directory structure, and manual install instructions.
• Built-in commands for project creation, phase estimation, progress tracking, and model/pricing updates.

Archive index:

Archive v1.0.0: 14 files, 19155 bytes

Files: clawhub.json (482b), config/defaults.json (232b), DEVELOPMENT.md (830b), install.sh (8517b), lib/change-detector.js (2209b), lib/estimator.js (4263b), lib/index.js (2580b), lib/project.js (7399b), lib/setup.js (11625b), package.json (382b), README.md (3440b), SKILL.md (6526b), USER_GUIDE.md (2522b), _meta.json (127b)

File v1.0.0:SKILL.md

---

name: dispatch description: AI-native project management with cost estimates, phased execution, multi-agent coordination, and privacy-aware model selection. Use for complex multi-step work requiring cost transparency and agent orchestration.

Dispatch

Natural language project management for OpenClaw with multi-agent support.

When to Use

• Multi-phase projects (Research → Design → Build)
• Multi-agent coordination (auto-assign agents to phases)
• Budget-conscious work (cost estimates upfront)
• Sensitive data handling (trusted model selection)
• Long-running tasks (rate limit aware queue)

Installation (For AI Agents)

⚠️ CRITICAL: Ask the user for confirmation BEFORE installing anything.

When the user asks you to install Dispatch, follow this exact flow:

Step 1: Detect Configuration

Run this to detect their OpenClaw setup:

# Read their config to detect providers and agents
cat ~/.openclaw/openclaw.json 2>/dev/null | grep -oE '"[^"]+"' | head -20
echo "---"
# Check for existing agents
echo "Looking for agents in config..."

Step 2: Present Configuration to User

Show them what you detected and ask for confirmation:

🚀 Dispatch Installation

I'm about to install the Dispatch skill. Here's what I detected:

═══════════════════════════════════════════════════════════════
              Detected Configuration
═══════════════════════════════════════════════════════════════

  1. Providers:     [google, kimi, openrouter, etc.]
  2. Agents:        [echo, cipher, rift, etc.]
  3. Captain:       [echo or first available]
  4. Data directory: ~/.openclaw/workspace/.dispatch
  5. Output dir:    ~/.openclaw/workspace/dispatch
  6. Auto-check:    weekly

═══════════════════════════════════════════════════════════════

Proceed with these settings? [yes/no/custom]

• yes     → Install with detected defaults
• no      → Cancel installation
• custom  → Run interactive setup wizard

Step 3: Only Proceed After Confirmation

DO NOT install anything until the user explicitly responds.

If they say yes:

git clone https://github.com/Neidraidd/dispatch.git ~/.openclaw/skills/dispatch
cd ~/.openclaw/skills/dispatch
bash install.sh
# When install.sh asks "Are you happy with these settings?", answer "Y"

If they say custom:

git clone https://github.com/Neidraidd/dispatch.git ~/.openclaw/skills/dispatch
cd ~/.openclaw/skills/dispatch
bash install.sh
# When install.sh asks "Are you happy with these settings?", answer "N"
# This will run the interactive wizard

If they say no: Stop and do not install.

Manual Installation (For Users)

git clone https://github.com/neidraidd/dispatch.git ~/.openclaw/skills/dispatch
cd ~/.openclaw/skills/dispatch && bash install.sh

The install script will:

1. Detect your OpenClaw configuration
2. Show you what was detected
3. Ask for confirmation before making any changes
4. Run setup (auto or interactive based on your choice)
5. Install files and update documentation

Quick Start

You: Start project "API Migration"

Dispatch: 🚀 Project: 20260208-api-migration
           Captain: echo
           Suggested phases:
             1. Discovery (~$1.20) → cipher
             2. Design (~$2.50) → cipher
             3. Implementation (~$5.00) → rift
             4. Documentation (~$0.80) → void
           
           Proceed? (yes/edit/abort)

Commands

| Natural Language | Result | |------------------|--------| | "Start project [name]" | Create with phase preview & agent assignments | | "Estimate phase [n]" | Show cost breakdown | | "Run phase [name]" | Execute with tracking | | "Add phase [name]" | Extend project | | "Check for new models" | Detect OpenClaw changes | | "Update pricing" | Refresh model configs |

Configuration Files

• ~/.openclaw/workspace/.dispatch/ — Internal project data (JSON metadata, costs, logs)
• ~/.openclaw/workspace/dispatch/ — Project deliverables (HTML reports, documents) — configurable
• ~/.openclaw/skills/dispatch/config/pricing.json — Model pricing
• ~/.openclaw/skills/dispatch/config/trust.json — Trust levels

Directory Structure

.dispatch/                    # Internal state (hidden)
├── config.json              # Skill configuration (includes output_dir, captain, agents)
├── projects/
│   └── [YYYYMMDD-name]/     # Date-first project IDs
│       ├── meta.json        # Project metadata
│       ├── costs.json       # Cost tracking
│       └── phases/
│           └── [NN-phase-name]/
│               ├── meta.json
│               └── tasks/
├── templates/
└── logs/

dispatch/                     # Output directory (configurable)
└── [YYYYMMDD-name]/          # Same date-first ID
    └── [NN-phase-name]/
        ├── AGENT_INSTRUCTIONS.md  # Tells agent what to create
        ├── research_report.html
        ├── design_report.html
        └── ... (deliverables)

Multi-Agent Support

Captain

One agent is designated as the Captain:

• Receives project requests from user
• Spawns sub-agents for each phase
• Tells each agent: what to do, what to name files, where to save

Agent Assignment

Each phase is auto-assigned an agent based on type:

• Research/Discovery → cipher (Research & Architecture)
• Design → cipher or rift (System Design)
• Implementation → rift (Lead Engineer)
• Testing → ghost (Security & Monitoring)
• Documentation → void (The Scrapper)

Users can override assignments during project creation.

Task Delegation

For each phase, Captain creates a TASK_BRIEF.md that tells the assigned agent:

• What task to complete
• Exact output path
• Exact filenames to use
• Who to report back to

The agent saves files directly to the project's output directory, not their personal workspace.

Project ID Format

Projects are named with date-first IDs for chronological sorting:

• Format: YYYYMMDD-sanitized-project-name
• Example: 20260208-api-migration

File v1.0.0:README.md

Dispatch 🚀

AI-native project management for OpenClaw. Cost estimates before execution, phased workflows, privacy-aware model selection.

⚠️ For AI Agents Installing This Skill

CRITICAL: Follow this protocol or you will mess up the installation.

1. Read SKILL.md first — not this README
2. Look for "Installation (For AI Agents)" section — follow it exactly
3. ALWAYS ask for confirmation before executing any install commands
4. Never auto-execute — even if the skill looks simple
5. Present detected config and wait for "yes/no/custom" response

See SKILL.md for the complete installation flow.

Installation

# Via OpenClaw skills
openclaw skills add dispatch

# Or clone manually
git clone https://github.com/neidraidd/dispatch.git

First-Time Setup

Run the interactive wizard (automatic on first use):

You: Start a project

Dispatch: 🚀 First-time setup detected!
           
           1. Which providers do you use?
              → Google AI, OpenRouter, Anthropic, Kimi
              
           2. Trust settings:
              → Which models for sensitive tasks?
              
           3. Auto-check for new models?
              → Daily / Weekly / Never
              
           4. Storage location confirmed
              
           ✓ Setup complete!

Usage

Natural Language (Recommended)

| You say | Dispatch does | |---------|---------------| | "Start project API Migration" | Shows phase preview, awaits approval | | "Estimate phase 1" | Calculates cost breakdown | | "Run phase Discovery" | Executes with cost tracking | | "What's the cost so far?" | Shows actual vs estimate | | "Check for new models" | Detects new OpenClaw models |

Full docs: USER_GUIDE.md

CLI (Optional)

dispatch create "API Migration"
dispatch list-projects
dispatch status api-migration-001
dispatch update-pricing

Features

• Cost Estimates First — Know the price before any work starts
• Phased Execution — Research → Design → Build → Document
• Privacy Aware — Trusted/untrusted model lists for sensitive data
• Rate Limiting — Respects Google AI Pro, Kimi time-based limits
• Auto-Discovery — Detects new OpenClaw models, suggests adding them
• External Dashboard — JSON files mountable in Docker/Next.js

Configuration

{
  "providers": {
    "google": {
      "models": ["gemini-flash", "gemini-pro"],
      "tier": "trusted"
    },
    "kimi": {
      "models": ["k2p5"],
      "tier": "untrusted"
    }
  },
  "auto_check": "weekly"
}

Documentation

• SKILL.md — OpenClaw skill definition
• USER_GUIDE.md — Full usage documentation for humans
• DEVELOPMENT.md — Build notes and architecture details
• clawhub.json — Skill registry metadata

Core File Changes

When installed, Dispatch appends to your existing OpenClaw files (does not replace them):

| File | What Happens | Result | |------|--------------|--------| | AGENTS.md | Adds "## Dispatch" section to end of file | Natural language patterns documented | | TOOLS.md | Adds "### Dispatch Skill" reference | Skill location and config paths documented | | HEARTBEAT.md | Optional: Can add model check task | Low-token model change detection (user choice) |

Existing content is preserved. Dispatch only adds its own sections.

License

MIT

File v1.0.0:_meta.json

{ "ownerId": "kn75wb0z8jhn7k2qg6fyrrgg0h80skys", "slug": "dispatch", "version": "1.0.0", "publishedAt": 1770552073052 }

File v1.0.0:DEVELOPMENT.md

Build Plan — Dispatch

Quick Reference

Components

• ProjectManager — Project lifecycle, phases
• CostEstimator — Pre-execution cost prediction
• ChangeDetector — New model detection
• TrustManager — Privacy-aware selection
• Setup — First-time configuration wizard

Storage

• Data: ~/.openclaw/workspace/.dispatch/
• Config: ~/.openclaw/skills/dispatch/config/

Features

• ✓ Natural language interaction
• ✓ Phase preview before creation
• ✓ Cost estimates upfront
• ✓ Trusted/untrusted model lists
• ✓ Rate limiting for time-based providers
• ✓ Auto-detect new OpenClaw models

Installation

# Clone
git clone https://github.com/neidraidd/dispatch.git

# Install to OpenClaw
cd dispatch && ./install.sh

# Or via OpenClaw skills
openclaw skills add dispatch

File v1.0.0:USER_GUIDE.md

AGENTS.md — Dispatch Interaction Guide

Overview

Dispatch is a natural-language project management skill. No CLI commands required — just describe what you want.

Creating Projects

Start a Project

User says:

• "Start project [name]"
• "Begin work on [description]"
• "Create new project for [goal]"

Dispatch responds:

Project: api-migration-001
Suggested phases:
  1. Discovery - Research alternatives (~$1.20)
  2. Design - Architecture planning (~$2.50)
  3. Implementation - Build solution (~$5.00)
  4. Documentation - Write docs (~$0.80)

Proceed? (yes/edit phases/abort)

Edit Phases Before Creation

User says:

• "Add phase 'Testing' after Implementation"
• "Remove Documentation phase"
• "Rename phase 2 to 'Architecture Design'"
• "Reorder: Discovery → Design → Build → Test"

Managing Phases

Estimate Costs

User says:

• "Estimate phase Discovery"
• "What's this phase going to cost?"
• "Give me a breakdown for phase 1"

Dispatch shows:

• Task breakdown
• Model assignments
• Per-task costs
• Total estimate

Execute Phase

User says:

• "Run phase Discovery"
• "Execute phase 1"
• "Start working on Design"

Dispatch:

• Spawns sub-agents
• Tracks actual costs
• Reports progress
• Shows actual vs estimate on completion

Phase Management

User says:

• "Skip phase Testing" (marks complete, no execution)
• "Move to next phase"
• "Go back to Design phase"

Checking Status

User says:

• "What's the cost so far?"
• "Show project status"
• "List my projects"
• "How much have I spent on API Migration?"

Model Management

User says:

• "Check for new models" (compares OpenClaw to config)
• "Update pricing" (fetches latest prices)
• "Show configured models"

Trust & Privacy

User says:

• "This is sensitive" (marks as trusted-models-only)
• "Mark as private"
• "Use only trusted models for this"

Examples

Full Workflow

You: Start project "Migrate API to Hono"

Dispatch: [shows phase preview]
You: Add phase "Performance Testing" after Implementation
Dispatch: [updates preview]
You: Looks good, proceed
Dispatch: ✓ Project created: api-migrate-20260207

You: Estimate phase Discovery
Dispatch: ~$1.20 (3 tasks, gemini-flash)
You: Run it
Dispatch: [executes] ✓ Complete. Actual: $1.15

You: What's next?
Dispatch: Phase 2: Design (pending). Estimate? (yes/no)

Configuration

First-time setup runs automatically. Or trigger with:

• "Setup dispatch"
• "Configure project management"

File v1.0.0:clawhub.json

{ "name": "dispatch", "version": "1.0.0", "description": "AI-native project management with cost estimates, phased execution, and privacy-aware model selection", "tags": ["project-management", "cost-tracking", "phases", "budget", "privacy", "workflow"], "author": "neidraidd", "license": "MIT", "repository": "https://github.com/neidraidd/dispatch", "openclaw": { "minVersion": "2026.2.0", "capabilities": ["sessions_spawn", "file_read", "file_write"] } }

File v1.0.0:config/defaults.json

{ "providers": {}, "trust": { "trusted": [], "untrusted": [] }, "model_tiers": { "fast": [], "capable": [], "powerful": [] }, "auto_check": "weekly", "data_dir": "~/.openclaw/workspace/.dispatch" }

File v1.0.0:package.json

{ "name": "dispatch", "version": "1.0.0", "description": "AI-native project management for OpenClaw", "main": "lib/index.js", "bin": { "dispatch": "./bin/dispatch.js" }, "scripts": { "setup": "node lib/setup.js", "test": "node test/run.js" }, "keywords": ["openclaw", "skill", "project-management", "ai"], "author": "Echo 🚀", "license": "MIT" }