lobster

What This Skill Does

Lobster is a powerful workflow orchestration engine designed for OpenClaw agents to manage deterministic, multi-step pipelines. It bridges the gap between raw AI planning and reliable execution by introducing structured data processing and mandatory human-in-the-loop approval gates. By defining tasks as typed JSON pipelines, Lobster ensures that sequences of commands—ranging from data transformation to external system interactions—are executed consistently and repeatably without requiring the agent to re-calculate logic at every single step. It is built to save tokens and improve reliability by treating complex workflows as immutable, stateful assets.

Installation

To integrate Lobster into your agent environment, use the OpenClaw installation command: clawhub install openclaw/skills/skills/guwidoe/lobster

Alternatively, you can manage it via Node.js by installing the package globally: npm install -g @clawdbot/lobster Once installed, map the binary path to your environment aliases to allow the agent to execute workflows efficiently using the $LOBSTER command.

Use Cases

Lobster is optimized for high-stakes or high-volume operations where precision is required. Key use cases include:

• Approval-gated Actions: Performing batch operations that require human review before executing side effects (e.g., sending emails or modifying database records).
• Stateful Monitoring: Tracking changes in external resources like GitHub PRs or Jira tickets, where you only want the agent to react when a specific change occurs.
• Data Pipelines: Cleaning, sorting, and transforming raw JSON outputs from shell commands into structured reports.
• Automated Triage: Sorting large volumes of tickets or emails through a chain of filters and projections before presenting a concise summary to the user.

Example Prompts

1. "Use Lobster to monitor the GitHub repo 'guwidoe/lobster-skill'. If there is a new PR, list the title and author, then wait for my approval before adding a 'triage' label."
2. "Fetch the list of merged PRs for the current repository, filter them by date, and format the output as a table."
3. "Run a data pipeline that pulls user records, filters for 'inactive' status, and stores the state so I am only notified when a user's status changes from active to inactive."

Tips & Limitations

• Tool Mode: Always prefer tool mode (--mode tool) when using Lobster within an agent workflow, as it provides a structured JSON envelope containing resume tokens for approval tasks.
• State Management: Leverage diff.last to keep workflows lightweight; it is the most efficient way to prevent spamming notifications or redundant actions.
• Limitations: Lobster is designed for deterministic pipelines. If your workflow requires high levels of heuristic decision-making or creative reasoning that changes mid-execution, stick to standard agent planning rather than Lobster pipelines.
• Error Handling: Ensure your shell commands return valid JSON, as Lobster's internal parser will fail on malformed output.