Collaboration & replay

The features in this section are about working with more than just one agent in isolation. Agents coordinate with each other and with you through the Coordination Bus. Finished sessions can be stepped back through and shared. Your best prompts live in a reusable library, and Prompt Copilot helps you write better ones. And when you sync with a team, everyone sees the outcomes without anyone sharing code.

Coordination Bus

The Coordination Bus lets agents talk to each other and to you. It carries inter-agent messages, a shared task graph, and decision gates that pause an agent until it gets an answer. The most useful of these is the ask-human gate: an agent can block on a question and route it to you, and your answer comes back as the result of its own tool call — so a running agent can genuinely stop and ask before doing something irreversible.

• Messages — agents can send each other direct or broadcast messages, including targeting whichever agent is idle.
• Task graph — a shared set of tasks agents can claim and mark done, so work is not duplicated across sessions.
• Decision gates — an agent blocks until the gate is resolved, including ask-human (only you can answer) and ask-session (another agent or you can answer).

Interactive Session History and Share

Every captured session can be stepped back through. This is not a screen recording — nothing is filmed. The player reconstructs the run from the captured trace data as an interactive timeline: traces on one side, the code and conversation on the other, with a scrubber, transport controls, and variable speed. Dead time between traces is compressed so a long session plays back quickly, and you can seek to any point.

You can also share a session history via a link, which is how you show a teammate exactly what happened without screen-sharing or copying logs. When you generate the link you control what it exposes and how long it lasts.

1. 1

   Open the share dialog

   From a session, choose Share. Operon prepares the shareable history from the session’s captured traces.

2. 2

   Set redaction toggles

   Choose what to strip before sharing — file contents, bash output, and tool inputs can each be redacted independently.

3. 3

   Pick an expiry

   Set how long the link stays valid — 24 hours, 7 days, 30 days, or never.

4. 4

   Generate and copy the link

   Operon creates the shareable link and copies it to your clipboard, ready to send.

Prompt Library

The Prompt Library, opened with Cmd+L, is where your reusable prompts live. It ships with a set of curated built-in prompts and holds the ones you save yourself, searchable and taggable. From the library you can send a prompt straight to the running agent, copy it, or edit it — and each prompt shows a quality badge so you can tell a strong prompt from a weak one at a glance.

Prompt Copilot

Prompt Copilot scores the quality of a goal as you type it and suggests improvements. It checks for the things that make a prompt reliable — concrete file references, clear action verbs, a stated outcome, defined scope, and a testing expectation — and surfaces a score plus targeted suggestions. It runs on the goal field when you start a new session, so a vague prompt gets nudged toward a good one before the agent ever runs.

Team observability

When you sign in and sync with a team, everyone gets a shared view of outcomes — decisions, sessions, cost, and progress. By default that sync also includes prompt text, agent responses, tool inputs/outputs, and code diffs; turn off the per-project code-content toggle if you want metadata only. The team layer is opt-in, and the work itself still runs on each developer's own machine — Operon never uploads your repository as files.

Related