arbe-delegate

Delegates coding outcomes to worker agents in herdr panes and collects verified reports.

Updated May 16, 2026
One-click install
npx skills add https://github.com/oskarrough/robots --skill arbe-delegate-oskarrough
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: arbe-delegate
Source: https://github.com/oskarrough/robots/tree/main/skills/delegate
Command: npx skills add https://github.com/oskarrough/robots --skill arbe-delegate-oskarrough

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Coordinating parallel AI coding agents is error-prone: workers start on the wrong model, prompts get lost, and results are hard to verify. This Skill turns one worker turn into one JSON envelope inside a live Herdr session, handling spawn, brief, settle confirmation, and result reading so you can review outcomes instead of babysitting panes. ## Core Features & Use Cases - One-call worker lifecycle: Spawn a fresh worker, verify its runtime (provider/model/thinking) against the request, send a brief with the reporting contract, and receive a structured envelope with classification, last_message, and terminal_text. - Warm worker steering: Send follow-up prompts or resume after timeouts with herdr-delegate prompt and herdr-delegate wait, with transcript-based settle confirmation that outranks flaky status signals. - Model selection guidance: A maintained models reference maps task types (implementation, review, scouting) to concrete --kind and provider/model arguments. - Use Case: You need a refactor implemented while you keep planning. Run herdr-delegate fix-auth 'BRIEF' --kind pi --timeout 120000 --new-tab workers -- --provider openai-codex --model gpt-6-astra --thinking low, then review the returned report and corroborate it with jj st. ## Quick Start Ask the agent to delegate one well-scoped coding outcome to a fresh herdr worker using herdr-delegate with a model from the models reference, then review the returned envelope's classification and last_message.

Frequently Asked Questions about arbe-delegate

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I delegate a coding task to a worker agent in herdr?▼

Run herdr-delegate with a unique name, a brief, a --kind, a timeout, and placement flags like --new-tab workers, followed by the agent's provider and model args. The wrapper spawns the pane, verifies the runtime, sends the brief, and returns one JSON envelope with the worker's report.

How do I choose which model to spawn for a delegated task?▼

Consult the models reference before every spawn: it maps task types like implementation, review, and scouting to concrete --kind and --provider/--model arguments. Pick based on task difficulty, since cheap models suit probes while strong models are required for reviews.

What should I do when herdr-delegate times out?▼

On a timeout at the prompt or wait stage, follow with herdr-delegate wait NAME --timeout <ms> and never re-prompt, since the worker may still be running. Budget more than the 20-second confirmation interval for each wait call.

Why does herdr-delegate report a runtime mismatch at the verify stage?▼

A herdr_delegate_runtime_mismatch means the worker started on a different provider, model, or thinking level than requested, detected by parsing the session transcript. Close that pane and start over rather than briefing the wrong-model worker.

Can I reuse a worker for follow-up tasks?▼

Yes, use herdr-delegate prompt NAME 'TEXT' to steer a warm worker without repeating the reporting contract. Reset context with a /new prompt before unrelated work, and reuse roughly four warm workers for batch jobs.

What are the limitations of herdr-delegate?▼

It requires a live Herdr session with HERDR_ENV=1 and never closes panes itself. Kinds without a known transcript format settle on status alone, and a killed worker loses its agent handle permanently, requiring a fresh spawn.