← ClaudeAtlas

grand-admirallisted

This skill should be used when coordinating multiple agents — spawning subagents, isolating worktrees, managing teams, scaling work, recovering failures, or managing a programme. Coordinator agents that spawn, manage, or merge subagent work always load it.
lklimek/claudius · ★ 1 · AI & Automation · score 71
Install: claude install-skill lklimek/claudius
# Grand Admiral — Multi-Agent Orchestration Operations manual for coordinator agents. ## Session Protocol - Load /git-and-github and /coding-best-practices at session start. Apply /coding-best-practices to ALL code work — yours and every agent's — continuously, not as a one-time read; require the same of every code-touching agent you brief (Agent Prompt Requirements #11). - Reread available skills and agents before each task - **Base freshness before planning** — check HEAD against the **base branch** it was cut from (`git rev-list --count HEAD..origin/<base>`), not against `@{upstream}`: a feature branch tracking `origin/<itself>` reads as current while the base moves PRs ahead, and that surfaces only as a merge conflict. Then act on what moved rather than on the count — read the base commits touching the files the task will touch (`git log -p HEAD..origin/<base> -- <path>`), and rebase or fold the overlap into the plan. You own this check even when the session-start probe said nothing — it is silent on detached HEAD, missing upstream, unresolvable base, and an unreachable remote, so silence is not evidence of freshness. - Check MemCan (if available): `memcan:recall` for architecture decisions, coding standards, design patterns, known pitfalls, and the user's mindset/values; `search_code` for existing implementations; `search_standards` for compliance. - **Search memory on surprise, not only at session start** — an unexpected result, a repeating error, or an approach fail