| name | khala-fleet |
| description | Operate an OpenAgents Khala coding fleet safely. Use when the user asks to connect or list fleet accounts, spawn/delegate coding work to Codex or Claude workers, start/monitor/pause/drain/stop a sustained fleet run, burn down a backlog with parallel workers, verify fleet work actually completed, or diagnose dispatch failures like "0/1 available" or target_pylon_unavailable. |
Khala Fleet Management
This is the canonical, repo-scope khala-fleet skill. It is a launcher, not
the law: the authoritative procedure lives in this repo and wins whenever
anything disagrees. Read these first, in full, every invocation:
- The Khala -> Pylon -> Codex Coding Delegation Runbook section of
AGENTS.md / CLAUDE.md at the repo root.
docs/khala-code/2026-07-02-khala-fleet-bundled-skill.md (how this skill is
bundled, discovered at repo scope, and materialized into
~/.agents/skills/khala-fleet/ by Khala Code Desktop).- The fleet spec under
specs/khala-fleet-delegate/ and the
packages/khala-fleet-intents/ intents package.
Quick orientation while you open those:
- Fleet = the owner's linked Codex/Claude accounts, each an ISOLATED local
worker reached through Khala -> Pylon -> assignment.
- Dispatch ladder: one bounded task (
codex_spawn / $PYLON khala request --workflow codex_agent_task with pinned repo/commit/verify) → parallel
wave (heartbeat capacity first, one claim per unit) → sustained fleet run
(fleet_run_start / fleet_run_status / fleet_run_control).
- Done = closeout checklist ok + exact
token_usage_events rows. Counter
movement alone is NEVER completion evidence.
- Never run login flows against
~/.codex or the owner's live ~/.claude.
worker auth uses isolated per-account homes only.
