skills.rest
midsphere-ai
midsphere-aiOFFICIAL
Explore all skills by this author →
WebsiteGitHub

exo:harness

Orchestrate event-driven agent workflows with custom harnesses and middleware.

1|Updated Feb 16, 2026
Search Tags:#middleware#agents#streaming#orchestration#swarm#checkpoint#harness
One-click install
npx skills add https://github.com/midsphere-ai/exo --skill exo-harness
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: exo:harness
Source: https://github.com/midsphere-ai/exo/tree/main/skills/harness
Command: npx skills add https://github.com/midsphere-ai/exo --skill exo-harness

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Orchestrate complex, event-driven agent workflows by building custom harnesses that coordinate multiple agents, intercept events in real-time, and manage per-harness state and history.

Core Features & Use Cases

  • Harness ABC and HarnessContext for orchestrating custom flows, with execute() as the entry point.
  • Middleware support (TimeoutMiddleware, CostTrackingMiddleware, and custom middlewares) to control timing and costs.
  • Parallel sub-agent execution via SubAgentTask, SubAgentResult, and SubAgentStatus for fan-out patterns.
  • Checkpointing and resume using MemoryStore and CheckpointAdapter to persist progress across restarts.
  • HarnessNode integration to compose harnesses within Swarms and power modular workflows.
  • Run/stream interfaces (run and run.stream) for non-streaming and streaming consumption of harness output.

Quick Start

Create a custom harness by extending Harness and implementing execute(), then run it with run(harness, input) to begin orchestration.

Frequently Asked Questions about exo:harness

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

FAQPage Schema
How do I orchestrate complex event-driven agent workflows with custom logic?▼

You orchestrate event-driven agent workflows by building custom harnesses that implement the execute() method to coordinate agents, intercept events in real-time, and manage per-harness state. This approach handles deterministic orchestration, real-time routing, and parallel execution across Swarms.

Can I run parallel sub-agent execution and stream the output?▼

Yes, parallel sub-agent execution is supported via SubAgentTask, SubAgentResult, and SubAgentStatus for fan-out patterns. You can stream output using the run.stream interface and stream_agents_parallel utilities within the HarnessContext.

How does checkpointing work for agent orchestration across restarts?▼

Checkpointing persists orchestration progress using MemoryStore and CheckpointAdapter to save and resume state. This allows long-running agent workflows to recover from interruptions without losing completed steps or per-harness history.

What middleware is available to control timing and costs in agent orchestration?▼

Built-in middleware includes TimeoutMiddleware to control execution timing and CostTrackingMiddleware to monitor spending. You can also add custom middlewares to intercept events and apply cross-cutting concerns across your harness workflows.

Does this orchestration approach work with existing Swarms?▼

Yes, HarnessNode integration allows you to compose custom harnesses directly within Swarms. This supports modular workflow designs where orchestrated agent flows operate as nodes inside broader swarm architectures.

What is the best way to start building a custom agent orchestration flow?▼

The best way to start is extending the Harness ABC and implementing the execute() method as your entry point. You then run the orchestration using run(harness, input) to begin coordinating agents and processing events.

Related Skills

Explore More Software Engineering Skills→
monoesmonoes
15

swarm-orchestration

Orchestrate multi-agent swarms for parallel task execution in Monomind workflows.

Official
Advanced
teegin-gteegin-g

databricks-spark-structured-streaming

Provide patterns and templates for Spark Structured Streaming pipelines with Kafka ingestion.

Community
Advanced
scooter-lacroixscooter-lacroix
8

hooks

Register TypeScript logic hooks for AI agent lifecycle events.

Community
Advanced
shaun0927shaun0927

parallelize

Coordinate parallel execution of independent tasks across multiple agents.

Community
Intermediate