HerdRanger 0.20.18 pre-release guide

04 · Agents and mobile workflow

Read the state. Respond once. Verify the result.

Check current Herdr state, review local captures, and choose how to respond.

Open the user guide

Find agents that need attention

In Core, open the left tree and choose Needs input. The count reflects available machines. In TUI, review the selected session’s agent status. Working indicates active work, Blocked or a waiting status indicates a request for input, and Unknown means you should verify the connection and target.

Standalone SSH gives you a shell without Herdr agent status or approvals.

Read live output and captured history

Core live output follows new content until you scroll back. The composer stays below the output. Long cleared terminal padding is condensed while ordinary spacing and fenced code are retained.

Open History to review acknowledged instructions and earlier captured output. Use the filters and source labels to identify the records, select text from captures, or choose Reuse in composer when available. The draft remains yours to review before sending.

Captured screens and any retained scrollback are bounded. History can show observed content still in retention. Consult the live agent or host for missing or discarded output. History is not a complete execution log and does not prove which agent changed a file.

For workspace differences, see read-only Changes. A delivery acknowledgment, an output capture, and a Git diff answer different questions: receipt, observed output, and changed content.

Review and send an instruction

Confirm the target and review the draft. Where a staged send offers Stop & edit or Send now, use those controls to revise or send the pending draft. A follow-up accepted behind active work can be marked Queued.

An acknowledgment confirms receipt by Herdr. It does not confirm the work finished. For uncertain delivery, check the target before retrying. HerdRanger preserves the draft and avoids automatic replays.

Respond to an approval request

Where the connected mode exposes reviewed approval controls, compare the target and the exact choice label with the current agent screen before confirming. If the request changed or the buttons are unavailable, review the current screen and use the appropriate reply or interactive terminal.

Provider tools and sign-in run on your host. You can use a Gemini agent configured there. This pre-release build excludes online prompt enhancement for drafts on your phone.

Open interactive controls deliberately

Viewing output is passive. Use the explicit live terminal action if the agent needs interactive sign-in, an MFA step, or terminal-only input. Confirm the exact target before typing. Esc can interrupt the remote work and requires confirmation where offered.

Planned activity tracking

Planned, not available yet. This describes the product direction, not controls in the current guide. No release date is announced.

The first planned timeline would bring together confirmed app operations such as instruction delivery, queue changes, approval responses, and interrupt requests. Clear source labels would distinguish a selected action from an acknowledged result.

Deeper tracking would show commands, outcomes, and affected files only when the host or provider supplies structured events with agent identity. Git differences and terminal text alone cannot establish authorship, especially when people and multiple agents share a project. Coverage gaps would remain visible.

The same work aims to make Changes easier to reach and delivery states easier to understand while keeping the existing chat, navigation, and visual style. See the public roadmap and the planned review workflow.