Replay 0.3.0: Muse Code support and real sub-agent coverage

Replay 0.3.0 is out. This release adds a new agent to the coverage matrix and fixes a gap that mattered more as agents started spawning other agents: child sessions were not always treated as separate sessions.

Muse Code is now supported for live monitoring, opt-in enforcement, and session rebuild from saved history. Sub-agent activity is also carried through capture, storage, timelines, finding drill-downs, and exports, so a child run no longer disappears into its parent.

Muse Code joins the matrix

Muse Code is a CLI agent, and Replay now covers the paths operators actually need:

• Session rebuild from Muse’s saved session transcripts.
• Live monitoring through Muse’s hook system, so tool calls and outcomes show up in Replay as they happen.
• Opt-in enforcement, where a matching rule marked for enforcement can block the action instead of only recording it.

Muse loads hooks when a session starts. After deploying protection for Muse Code, start a new Muse session before expecting live capture. The Agents view will show the connector as wired once the configuration is in place; Verify runs a synthetic action through the installed hook to confirm a record actually comes back.

As with every other agent, Replay stays in monitor mode until you deliberately enable enforcement. Installing the connector is not the same as turning on blocking.

Sub-agents are sessions too

A parent agent that spawns a reviewer, a tester, or a second coding thread can produce several sessions that share the same agent and even the same session identifier. Replay used to group sessions only by agent and session ID. That was enough while child identity was invisible, but it broke down once agents started reporting it. Two child sessions could overwrite each other in the local store, collapse into one row in the UI, or send a finding’s drill-down to the wrong transcript.

Replay 0.3.0 keeps the identity the agent reports:

• session tree identifier
• parent session identifier
• sub-agent label
• stable sub-agent ID, when the source provides one

That identity now follows the session through storage, live grouping, timeline lookup, and the app’s session list. Child sessions that share an agent and session ID stay separate. A finding that points at a child session opens that child’s timeline, not the parent’s or a sibling’s. Exported session reports carry the same parent and sub-agent context.

Sources that do not provide a stable child ID still get their existing sub-agent label preserved. Replay does not invent lineage that the agent did not report.

What it looks like in the app

In Agents, Muse Code appears alongside the other covered agents with its own session-rebuild, live, and enforcement status.

In Sessions, a parent and its children can share a session ID without collapsing into one row. Each row carries the sub-agent label, and the detail pane shows the child ID and parent session when they are available.

In Alerts, session drill-down uses the finding’s sub-agent identity. If the alert came from a child run, the timeline that opens is the child’s.

Existing sessions still load. A new scan adds child sessions under their own identities; there is no migration step.

Upgrade notes

Replay 0.3.0 is a free update for existing installs. Download the current build for Linux, Windows, or macOS from the download section.

After upgrading, re-run the connector install for the agents you rely on, using the same policy, emit, enforcement, and delivery settings. That refreshes the generated hook definitions. For Muse Code, deploy protection and start a new Muse session before expecting live capture.

Sub-agent coverage depends on what the source reports. When an agent exposes child identity, Replay preserves it end to end. When it does not, Replay keeps the session intact and does not guess at relationships.

Ecliptica Replay is free and local-first for desktop, CLI, IDE, and gateway agents.