A deterministic workflow runtime for AI agents designed to execute typed JSON pipelines with mandatory human approval gates for side effects.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install lobster
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install lobster using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Lobster?
Lobster is a specialized runtime engine designed to bring reliability and safety to AI agent workflows. By utilizing deterministic pipelines rather than relying on an LLM to re-plan every individual step, Lobster minimizes token usage and prevents unexpected behaviors in complex automations. It is an essential component for developers building Openclaw Skills that require a human-in-the-loop for sensitive operations like modifying repositories or sending communications.
The tool operates by chaining commands together, allowing data to flow through structured stages including shell execution, JSON filtering, and stateful comparisons. This ensures that agents can perform batch operations, monitor external events, and maintain a consistent state across different execution cycles with high precision.
Lobster Use Cases
- Running multi-step automations that require manual verification before performing side effects.
- Monitoring GitHub Pull Requests or issues for specific changes and notifying stakeholders.
- Processing large batches of data through typed JSON pipelines for triage or reporting.
- Managing persistent state across repeated AI agent tasks to track progress or diffs.
- Integrating existing CLI tools into structured AI agent workflows with predictable output.
How Lobster Works
- The runtime receives a pipeline string or a .lobster workflow file containing a sequence of commands.
- Initial data is often gathered using shell commands via the exec tool, which parses output directly into JSON.
- Data flows through transformation pipes such as where (filtering), pick (projection), and sort to reach the desired state.
- If an approval gate is reached, the process halts, persists its current context, and returns a unique resume token.
- Once a human provides approval via the resume command, the workflow continues with the authorized data to complete the operation.
Lobster Setup
To begin using this tool within your Openclaw Skills environment, follow these installation steps:
# Install globally via npm
npm install -g @clawdbot/lobster
# Or set an alias to the local source if running from a specific directory
LOBSTER="node /path/to/lobster/bin/lobster.js"
# Verify the installation and list available commands
lobster commands.list
Lobster Data Schema & Taxonomy
Lobster manages data through structured JSON envelopes and local file-based state persistence. It organizes information as follows:
| Data Type | Description | Default Location |
|---|---|---|
| Pipeline State | Persistent key-value pairs used for diffing and memory | ~/.lobster/state/ |
| Workflow Definitions | YAML or JSON files defining reusable multi-step logic | User-defined (.lobster) |
| Approval Tokens | Base64 encoded strings containing the serialized state for resumption | CLI Output (JSON) |
| Tool Envelopes | Standardized JSON wrappers for AI agent tool calling | CLI Stdout |
Lobster Advanced Features
- Stateful differencing using diff.last to only trigger actions when data changes between runs.
- Seamless integration with the Clawdbot ecosystem via clawd.invoke for triggering external tools.
- Support for complex template rendering using double-curly brace syntax within pipelines.
- Local state management using state.get and state.set for long-running monitoring tasks.
- Tool mode vs. Human mode switching to provide either machine-readable JSON or pretty-printed tables.
SKILL.md
Loading
Related Openclaw Skills
Luban CLI
A comprehensive command-line framework for managing the lifecycle of machine learning experiments, training jobs, and inference services.
Gas Price Alert
An automated monitoring tool that tracks local gas prices and provides daily alerts for the cheapest fuel options including Costco.
Quick Test
A lightweight diagnostic utility to verify OpenClaw environment stability through automated system checks and command validation.
Personal Assistant
A minimalist productivity and well-being assistant that generates personalized daily briefings and habit tracking routines using Openclaw Skills.
Weather
A lightweight skill for retrieving real-time weather data and forecasts directly from the CLI without requiring API keys.
Feishu Contacts Search
A powerful directory search tool for Feishu that enables AI agents to locate users and departments via fuzzy matching and local caching.
Featured*
