OSA: bringing Agent-to-Agent collaboration into the developer workspace

Multiple computers. Multiple agents. One place to collaborate.

Running several coding agents raises a practical question: how do you keep track of their work? One agent is implementing a change, another is investigating a failure, and a third is waiting for a decision. Their conversations live in different windows. The work may also live on different computers.

Orca-superagent (OSA) brings that coordination into Orca through a Router skill, executor rules, and a native task panel.

Four ways work moves between people and agents

Human-to-Human collaboration runs on conversations, assignments, and handoffs. Human-to-Agent interaction lets a person describe a goal and receive an implementation. Agent-to-Human interaction brings questions and progress back to the person. Agent-to-Agent collaboration connects those loops: agents delegate work, return results, and pass questions to the right place.

OSA makes that fourth interaction part of the development workspace. The Router clarifies the intended outcome and routes the original request. Executors plan, investigate, implement, and debug. Reports and questions return to a shared Run inbox. The owner sets direction and makes the decisions that need their judgment.

Why an Orca plugin?

A task is easier to follow when its overview is connected to its execution session. OSA's sidebar organizes tasks into Needs you, Ready, Waiting for external events, In progress, and History. Tasks can be grouped by project, workspace, or agent.

Expanding a task brings up conversations and records. A task reference can be copied with @ and brought back to the Router. The panel gives users a path from “what is happening?” to the session where they can follow up.

Orca supplies the execution windows, workspaces, native orchestration objects, remote access, and multi-computer management. OSA supplies the coordination workflow and task view. Developers can access local and remote work environments through the same workbench while using agents such as Claude Code and Codex.

A concrete handoff

Describe a result to the Router:

Fix this project's login failure and verify the result locally.

The Router clarifies the target and dispatches an executor. The executor investigates the failure, plans the change, implements it, and reports the outcome. If a decision is needed, its question returns through the coordination workflow. Users follow the work from the panel and continue the conversation with its task context.

Long-running tasks also need review. OSA distinguishes questions for a person from waits for external conditions. A waiting task carries a review date; coordinator queue reviews bring it back into consideration. The project is evaluating the complete “A waits for B, B finishes, A continues without another human prompt” workflow using real tasks.

Try it

With Orca and a signed-in coding agent ready:

1. Put skills/osa-router/SKILL.md in your agent's skills directory.
2. Put skills/executor-rules.md at ~/osa-team/executor-rules.md.
3. From the repository root, run:

orca plugin install ./plugin/osa-panel

Then start in an Orca window:

/osa-router Fix the login failure in this project and verify it locally

The README is available in Chinese, English, French, Spanish, Russian, and Arabic. OSA draws on dsh-superagent's role separation and autoloop's event-and-clock approach, built on Orca's native workspace and orchestration capabilities.

Explore OSA · English README · Feedback

OSA code and documentation use PolyForm Noncommercial 1.0.0; Orca-derived patches use MIT.

Project introduction prepared with AI assistance from the maintainer's brief and repository documentation.