It depends: herdr can restore the layout and some supported agent sessions, but it cannot guarantee that the original Claude Code or Codex processes survive a reboot.

• The README states that after a server or machine restart, herdr restores the saved layout and “can resume supported agent sessions.”
• The same statement explicitly says, “the original processes do not survive,” so this is not process-level checkpointing or guaranteed task continuity.
• Recovery depends on whether the agent is supported and whether the relevant session state was saved. The README lists Claude Code, Codex, Cursor, OpenCode, and Grok, but does not define recovery coverage for each one.

The project therefore fits a requirement to recover the workspace view, but not a requirement that the exact original process continue executing without interruption.

Yes, because the README explicitly says that herdr does not wrap or replace existing agents; it owns the terminals in which they run.

• “runs what you already run” lists Claude Code, Cursor, OpenCode, Codex, and Grok, indicating that the intended workflow is to host existing tools.
• The README says, “herdr doesn’t wrap or replace them; it owns their terminals,” so migration means placing agents in herdr panes rather than rewriting their invocation model.
• It also provides tmux-style keyboard controls plus mouse clicking, dragging, and splitting, so an existing terminal workflow does not have to become entirely graphical.

However, generic terminal hosting does not guarantee full status detection, session recovery, or agent-native control for every agent. Those capabilities depend on the integration support available for the specific tool.

Yes, because agent-native control is an explicit design goal of herdr, and the README provides dedicated integration entry points.

• The “agent-native” section says agents can drive herdr through the CLI and Socket API rather than merely running inside its terminals.
• It explicitly lists spawning panes, prompting other agents, and waiting until another agent is genuinely blocked, which directly matches the requested orchestration behavior.
• The documentation index includes “agent skill,” “add herdr support,” and “socket api,” showing that agent integration is a first-class interface rather than keyboard automation alone.
• The plugin system can also extend panes and workflows.

The README does not provide the API schema, authentication model, error responses, compatibility policy, or safeguards against circular waits. Those omissions prevent a conclusion about production-grade orchestration readiness.

Yes, because herdr separates the lifetime of the terminal workspace from the lifetime of the client connection, specifically targeting SSH disconnects and parallel agent management.

• The README section “detach without stopping work” says terminals continue running in a background server after the client closes or the SSH connection is lost.
• “several machines, one window” places local work and saved SSH machines in one interface, with independent reconnects.
• The project explicitly supports Claude Code and Codex, and marks panes as working, blocked, or idle so you can find agents that need attention.

It does not remove failures caused by an unreachable host, invalid SSH credentials, or insufficient remote resources. The README does not specify reconnection behavior, supported SSH configurations, or resource limits when several agents run simultaneously.

Yes, because herdr preserves tmux-style keyboard interaction while adding mouse controls and visible agent states inside panes.

• The README section “keyboard and mouse, both first-class” explicitly supports tmux-style prefix keys as well as clicking, dragging, and splitting, so the keyboard workflow is not removed.
• “never hunt for the stuck one” says every pane is marked working, blocked, or idle, and alerts the user when an agent stops and needs an answer.
• The project is a terminal runtime and workspace manager rather than an Electron desktop application; the README emphasizes “one rust binary, no electron.”

However, a state label does not guarantee accurate detection for every agent. The project insight notes that output formats and interaction modes can affect blocked-state detection, while the README does not document per-agent detection rules or false-positive handling.

It depends: the delivery model is well aligned with consistent deployment across platforms, but endpoint-protected Windows requires separate validation.

• The README describes herdr as “one rust binary, no electron” and says it runs in the terminal you already use, reducing Electron runtime dependencies.
• The installation section provides shell installation, Homebrew, mise, a Windows PowerShell command, and binary downloads.
• It also links to a separate “endpoint-protected Windows” path; the project insight indicates that this Windows Beta route may have enterprise compatibility or deployment restrictions.
• Project data identifies Rust as the main language, with the overwhelming majority of the code in Rust, suggesting a relatively uniform delivery architecture.

It is therefore a good fit for ordinary cross-platform installation, but the README does not establish that it can be deployed directly where remote scripts, execution policies, or unverified binaries are restricted.