task-workflow

What This Skill Does

The task-workflow skill provides a rigorous, standardized framework for orchestrating complex, multi-stage operations within OpenClaw. Designed specifically for tasks that require more than five tool calls, extensive sub-tasking, or long-running execution times, this skill ensures that large-scale AI operations do not lose context or efficiency. It enforces a structured life-cycle consisting of initial planning, parallel subagent dispatch, automated progress tracking, and final integration. By utilizing a central task_plan.md file, the agent maintains a persistent, human-readable state, reducing the likelihood of hallucination or drift in long-running projects.

Installation

To install this skill, use the following command in your OpenClaw terminal: clawhub install openclaw/skills/skills/charpup/task-workflow Ensure your workspace has the necessary permissions to write files to the local directory, as this skill relies heavily on the generation and modification of the task_plan.md and MEMORY.md files.

Use Cases

This skill is intended for high-complexity workflows where standard linear execution is insufficient. Common use cases include:

1. Research projects requiring cross-referencing multiple datasets where individual subagents can handle different sources in parallel.
2. Complex software engineering tasks requiring a planning phase, code generation, testing, and documentation steps.
3. Long-running automation tasks that exceed 10 minutes of execution time, where periodic heartbeat checks are required to ensure system stability.

Example Prompts

1. "I need to research the historical impact of the 2008 financial crisis on three specific regions; please initialize the task-workflow and assign subagents to gather data from each."
2. "Execute the task-workflow for building a full-stack dashboard for my CRM data, ensuring you create a task plan and provide updates every five minutes."
3. "Start a complex workflow to audit my repository's dependencies; split the workload into small batches and handle the reporting via subagents."

Tips & Limitations

• Always ensure the task_plan.md file is accessible in the root of your project; the automated cron jobs depend on reading this file for progress updates.
• The skill is most effective when tasks can be decoupled; try to split your workload into independent, non-blocking batches to maximize the efficiency of parallel subagent dispatching.
• Monitor the "Risk & Blocking" section of your generated reports closely; the system is designed to highlight blockers that require human intervention, but the agent will only flag what it is explicitly programmed to detect.