Guide · Agent skill
How to use the Worker Delegation Playbook agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Worker Delegation Playbook is an agent skill from the Hive framework that teaches an AI agent how to split a large batch of similar work, such as many rows, accounts or files, across several parallel worker agents. It also covers when splitting is not worth it. This guide explains what the skill changes in the agent's behaviour, how to install it, how to trigger it, and what it needs before you rely on it. After reading, you should be able to decide whether your job suits this pattern and set up the skill correctly.
Worker Delegation Playbookaden-hive/hive · Apache-2.0 Download the skill Quick start
- 1Confirm colony mode and tools. Check that you are running the Hive framework in colony mode with the tools the skill names, such as run_playbook and tracker_sql.
- 2Install the ZIP. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills.
- 3Describe a uniform batch. Ask the agent to process a set of similar units, such as rows, accounts or files, in parallel.
- 4Let the agent build the tracker. The agent creates a table with one row per unit and a column that shows when each row is done.
- 5Approve the pilot. Let the agent run one representative row itself and patch the worker procedure if it hits a problem.
- 6Run the playbook. The agent writes and runs the playbook, which dispatches workers for undone rows and retries the gaps.
- 7Resume if interrupted. If the run stops, ask the agent to re-run the playbook by name; finished rows are skipped.
What the Worker Delegation skill makes the agent do
The skill applies when an agent is in Hive's colony mode and is considering whether to send work to parallel workers using the run_playbook tool. Instead of reading worker reports and deciding each next step by hand, the agent models the goal as a tracker table where every unit of work is a row. It then writes a playbook, a deterministic Python script, that repeatedly queries the rows that are not done, dispatches one worker per row, and stops when none remain.
The skill uses three artifacts. The tracker table holds the work list and its state. A worker skill, written with write_skill, holds the reusable procedure each worker follows. The playbook, run with run_playbook, handles orchestration such as which rows are undone, how many workers run at once and the retry policy. Because state lives in the table, re-running the playbook resumes the job and skips finished rows.
Install the skill on Claude
You can install the skill in one of two ways. The first is to download the ZIP from this page and unzip it into ~/.claude/skills/ for your user account, or into a project's .claude/skills/ folder to limit it to that project. The skill's files include a SKILL.md file.
The second is to upload the ZIP in the Claude apps under Settings, then Capabilities, then Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/ folder
- App install: upload the ZIP under Settings → Capabilities → Skills
Check the requirements before you trigger it
The skill only makes sense inside Hive colony mode. The instructions refer to specific tools, so the agent cannot follow them as written in a setup that lacks them. Required tools include run_playbook, write_skill, tracker_sql, tracker_register_writable and converge. Depending on the task, the worker procedures will also call browser or API tools such as hive-browser or web_scrape.
The skill runs scripts and uses the network, because the playbooks it produces are Python scripts that call workers that perform web or API work.
Trigger the skill with natural requests
You do not need a special command. Ask for a batch of similar items to be processed in parallel, and the agent should consider this skill when it is in colony mode. Good triggers name a large, uniform set of units, for example a list of profiles, records or files that each need the same several steps.
Be explicit about resumability if it matters. The skill is built around the idea that an interrupted run can pick up where it stopped.
- Process these 300 account pages in parallel, one worker per account, and resume if the run stops.
- Enrich each of these lead records using the same browser steps, in parallel.
- Split this file list across workers and keep a table of which files are finished.
Know when the skill says not to fan out
The skill is explicit that fan-out is not always worth it. Spawning a worker has overhead, since each one starts a fresh agent loop with a separate conversation and no shared context. With only one or two cheap units, or with exploratory work where the scope is still unclear, the agent should do the work itself or first break the task into bounded parts.
The skill also tells the agent to pilot one row itself before running workers, especially for dynamic web pages. Expect the agent to test the procedure on a single representative row first.
What people use it for
- Filling in profile fields for a long list of people or companies from web pages
- Processing many accounts or records with the same browser or API steps
- Working through a large set of files where each file can be handled on its own
- Resuming a partly finished batch job without repeating completed rows
Frequently asked questions
How do I install the worker delegation skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for your user, or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What do I need to run this skill?
You need the Hive framework in colony mode with tools such as run_playbook, write_skill, tracker_sql, tracker_register_writable and converge. Depending on the task, you also need browser or API tools such as hive-browser or web_scrape.
Is the worker delegation skill free?
The source licence is Apache-2.0. This page does not state pricing, so check the licence and the Hive project for any cost terms.
Does the skill run scripts and use the network?
Yes. The skill runs scripts and uses the network, because the playbooks are Python scripts that dispatch workers which perform browser or API work.
When should I not use this skill?
Do not use it for simple tasks with only one or two items, because spawning workers has overhead that outweighs the benefit. It is also a poor fit for open-ended exploratory work.
What happens if my run is interrupted?
Re-run the playbook by name. It queries the rows that are not yet done and skips completed ones, so you do not need to find the gap manually.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers and operators running multi-agent batch work on the Hive framework, especially when items are uniform and resumability matters. If you are using a different agent setup, or your job has only a few items, the skill is either unusable or not worth the overhead, so choose a simpler approach.
Download the skill