cord-trees

What This Skill Does

Cord-trees is a sophisticated orchestration skill for OpenClaw that empowers the agent to act as a project manager. Inspired by the Cord protocol, this skill shifts the paradigm from rigid, pre-defined automation scripts to dynamic, runtime-adaptive task trees. Instead of following a linear path, the agent analyzes the user's intent, breaks the high-level goal into logical components, and manages the lifecycle of these sub-tasks independently.

The core of this skill lies in its five powerful primitives: SPAWN (for isolated, clean-slate tasks), FORK (for context-aware synthesis), ASK (for human-in-the-loop checkpoints), SERIAL (for ensuring strict order), and the top-level GOAL. By managing its own task tree, the agent minimizes developer overhead and maximizes the intelligence applied to complex multi-step objectives.

Installation

To add this capability to your agent, run the following command in your OpenClaw terminal:

clawhub install openclaw/skills/skills/moltonbot000/cord-trees

Ensure your agent has the necessary permissions to read and write to the local file system, as the state tracking relies on a cord-state.json file to maintain the tree hierarchy across agent restarts or deep reasoning cycles.

Use Cases

Cord-trees is best suited for complex, non-linear workflows that involve ambiguity or multi-stage decision-making. Excellent use cases include:

• Researching technical migrations that require auditing current states and comparing future alternatives.
• Coordinating complex content production, where drafting, reviewing, and publishing must occur in specific sequences.
• Large-scale data processing tasks that require initial data retrieval, followed by conditional analysis based on the output of the first stage.
• Multi-agent orchestrations where the agent needs to break a project into chunks that can be distributed across various sub-agent contexts.

Example Prompts

1. "Decompose the task of migrating our documentation from Markdown to a database-backed CMS; figure out how to do it efficiently using cord-style task trees."
2. "I need a self-organizing agent plan to build a web scraper. Please build a task tree for the entire process, including auditing the target site and handling potential rate limits."
3. "Evaluate the performance of our current AWS Lambda infrastructure; act as an architect and build a coordination tree to identify bottlenecks and propose cost optimizations."

Tips & Limitations

• Context management: Use SPAWN for tasks where clean isolation is required to prevent 'context bloat' in sub-agent prompts.
• Human-in-the-loop: Always use the ASK primitive when the project reaches a point of high risk or when your business logic requires specific human approval before proceeding.
• Transparency: The agent's ability to 'self-organize' is only as good as the instructions provided. Always start with a high-level, clear goal to ensure the root node is well-defined.
• Limitations: This skill is not intended for high-frequency, millisecond-latency operations. It is designed for human-speed orchestration where thoughtful, structured reasoning is prioritized over raw execution speed.