fleet
The Broomva fleet: every Claude Code and Paseo session, PR and loop the coordinators drive, and the workflows that manage it. reconcile is the hourly launchd tick (observe, classify, report, owner ask
SKILL.mdskills/orchestration/fleet/SKILL.md
GitHubfleet
The fleet is every session, PR and loop that Broomva's standing coordinators
drive: the broomva coordinator (coordinator-tick), the SRI coordinator
(sri-coordinator-tick, its own scope), and the hourly reconcile tick that
reads them all. This skill is the entry point. Each thing done to the fleet
is a workflow with its own trigger, steps and a check you can run.
Launching is not this skill; managing is. Each peer gets its own worktree
→ bstack wave dispatch; peers share one worktree → bstack fleet up;
manage what's running → fleet reconcile (and census, close-out, rollover).
bstack fleet is bstack's launcher, a different command from this skill's
scripts/fleet; this skill puts nothing named fleet on PATH. Inside T3 Code
the launch row in workflows/launch.md sends peers through
T3 (delegate_task, t3_thread_launch) instead of either bstack launcher.
Named fleet-reconcile until 0.6.0 (BRO-3122). fleet-reconcile stays a
deprecated alias for one release: skills/orchestration/fleet-reconcile/
forwards its scripts here, which needs fleet installed beside it (as in
the repo; a per-skill install adds both:
npx skills add broomva/skills --skill fleet). The runtime names keep the old name (launchd
com.broomva.fleet-reconcile.<scope>, ~/.local/state/fleet-reconcile/,
~/.local/share/fleet-reconcile/releases/, the loop rows), because Maestro
and the loop ledger read them.
Workflows
| Workflow | Trigger | Mechanism | File |
|---|---|---|---|
| reconcile | launchd, hourly per scope; or fleet reconcile by hand |
code: scripts/tick.sh and the fleet CLI |
workflows/reconcile.md |
| census | every coordinator fire, before any launch, resume or archive | procedure: bstack base §4; the overlay's snapshot | workflows/census.md |
| close-out | a driven session finished; an owner or other-provider session the snapshot marks archive (base ≥ 0.48.0: owner sessions are driven too) |
procedure: bstack base §5–§6 | workflows/close-out.md |
| rollover | a driven session over 600k context; the coordinator over 400k | procedure: bstack base §11; the overlay's Rollover | workflows/rollover.md |
| launch | N peers to start: own worktrees, or one shared | routes to bstack: bstack wave dispatch, bstack fleet up |
workflows/launch.md |
Read the workflow file before acting. For census, close-out and rollover the
canonical text is bstack's references/orchestrator-tick.md, read from git;
the workflow file names its trigger and its check and links there rather
than copying it.
A fleet ask written by the reconcile tick is an acknowledgement with one
option, so it needs nothing beyond its headline. An ask a coordinator writes by
hand into .control/asks (a close-out choice, a policy question) follows the
decision layer: once it has three or more options, or is about a spec or a
design, it carries a summary, a tradeoffs row per option and the sources
it rests on, such as the PR, the ticket and the sessions involved (decision layer
v2, .control/asks/DECISION-LAYER.md; the rubric is in handback).
The CLI
F=skills/orchestration/fleet/scripts/fleet # or the pinned copy launchd runs
$F --help # every command; the workflows listed at the end
$F workflows # name, trigger, file
$F reconcile --scope broomva # one reconcile tick now (tick.sh, DRY_RUN=1)
$F rollover # print a procedure workflow (also census, close-out, launch)
$F asks --scope broomva # the reconcile tick's own verbs, unchanged
The module names stay fleet_reconcile.py and fleetlib/: they are the
reconcile workflow's implementation, and tick.sh, install.sh, the tests and
every pinned release name them.
Tests
cd skills/orchestration/fleet
python3 -m pytest tests/ -q # test_layout.py: each routed workflow exists; the alias resolves
python3 tests/mutation_check.py