drover_

LOCAL AGENTS / SHARED REPO / HUMAN AUTHORITY

A team of coding agents.
One accountable loop.

drover is an MIT-licensed macOS toolkit for developers who run teams of coding agents in parallel in Herdr, with task dispatch, isolated Git worktrees and cross-harness review.

MIT licensed · macOS · Bring your own coding CLIs

Know the boundary. Drover is local orchestration, not a model, hosted IDE or security sandbox. Agents use your credentials and native permissions. Inspect the dry-run and configuration before enabling automation.

01 / THE OPERATING LOOP

Keep work moving.
Keep proof attached.

01

Assign a task

One owner, a written scope and a Git worktree for the coding seat.

02

Build and test

The agent produces commits and executable evidence, not a status label.

03

Review across harnesses

Route the named change to a reviewer on a different coding harness.

04

Prove it landed

Apply the configured gate and verify the commit is on fetched main.

Read the operating model ↗

02 / FIELD NOTES, NOT PROMISES

The failures that
shaped the guardrails.

Historical reports from one operator’s fleet. Raw ledgers are not distributed; these are not independent benchmarks or promised improvements.

2026-09-07

44 min

Median dispatch gap

650 gaps; p90 285 minutes. Tasks took 10–20 minutes while the lead polled in eight-minute sleeps.

Read the dated lesson ↗

2026-09-07

10 seats

“Done” with no work

After credits ran out, ten agents reported done with no commit, report or error. Artifacts became the completion evidence.

Read the dated lesson ↗

2026-09-18

$952

The daily-meter error

A daily report counted lifetime session cost. The guard changed to per-message cost since local midnight.

Read the dated lesson ↗

03 / CHOOSE YOUR WORKFLOW

drover vs alternatives

First-party descriptions read 2026-10-19. No performance ranking is implied.

Documented coordination models
ProjectWhat its own documentation describes
droverA local Herdr fleet loop with dispatch, budget/ownership evidence, cross-harness review and gated publication. Operating model ↗
herdr-synapseA Herdr plugin with a shared team board, human-owned charter and idle-gated message delivery. First-party source ↗
Claude SquadA terminal app using tmux and Git worktrees to run coding agents on separate tasks and review changes. First-party source ↗
ConductorIts official docs describe parallel coding agents in isolated workspaces, with diff review and pull-request workflows. First-party source ↗
Vibe KanbanIts README describes kanban planning, agent workspaces and inline diff review; it also announces the project is sunsetting. First-party source ↗
Claude Code agent teamsAnthropic documents a lead session coordinating Claude Code teammates through shared tasks and inter-agent messages. First-party source ↗

Conductor has no product README in its public release repo; official docs are cited. Claude Code agent teams have their own official feature documentation. Claim-by-claim source audit ↗

04 / QUESTIONS YOU ACTUALLY ASK

Run it with eyes open.

How do I run multiple Claude Code agents in parallel?

Install Herdr and authenticate your coding CLIs, inspect Drover’s install dry-run, configure a lead plus separate coder/reviewer seats in examples/config, then run fleet up. Each assigned coder works in its own Git worktree. Start with two seats and one task; inspect the result before enabling watcher automation.

How do I stop coding agents double-dispatching?

Choose one assignment authority. Drover projects native task ownership and the fetched main board before dispatching; fleet-claims-audit reports disagreement as UNKNOWN. Check the task and artifact evidence before releasing an owner. A stale pane or a done status alone is not proof of completion.

How do I set up cross-harness code review with Codex and Claude?

Configure coder and reviewer seats on different harnesses, for example a Codex coder and a Claude reviewer. Drover routes the named commits for execution of the build and affected tests. Keep the same-harness review guard enabled; a prose approval is not evidence that tests ran.

Is Drover a model, hosted IDE or security sandbox?

No. It is local orchestration around existing CLIs in Herdr. You supply their accounts and your project’s build/test commands. Worktrees separate branches, not operating-system permissions. Review automation and native CLI permission settings before running it.

Does Drover support Linux or Windows?

The full operating loop targets macOS: launchd, BSD utilities and APFS clone operations are assumed. Portable individual scripts do not establish support for the complete fleet on Linux or Windows.

START SMALL

Two seats. One task.
Inspect the first loop.

git clone https://github.com/andrebrov/drover.git
cd drover
./install.sh --dry-run

Read the installation guide before installing or loading launch agents.

Installation guide ↗