agent-armylisted

Use when a large body of work is already split into independent groups (epics/stories, OpenSpec changes, migration sites, review dimensions) and you want to run many autonomous Claude agents in parallel — each isolated in its own git worktree and branch — coordinated by a changeable master session through a shared command centre that lives outside the repo. Triggers include /agent-army, "agent army", "fan out agents", "spin up a fleet", "parallel implementation streams", and long multi-stream operations that must survive context limits and shift changes. ALWAYS run the Front Door below first — it routes cheap work away from the fleet.
amjad1233/claude-skills · ★ 1 · AI & Automation · score 67

Install: claude install-skill amjad1233/claude-skills

# Agent Army 🫡 ## 🪖 Overview Run a **fleet of autonomous Claude agents in parallel**, one per independent group of work, each in its own git worktree/branch, coordinated by a **master** session through a **command centre** that lives *outside* any repo (in `~/.claude/agent-army/<operation>/`). The command centre is the single source of truth: the master and agents read/write it, and **any session can take command** by reading it — so the operation survives context-window limits and shift changes. **Core principle:** the master never trusts a single signal. Self-reported status *plus* objective health (process alive + last commit + transcript activity) together tell the truth. Agents stall silently; commits lag reports; launches lie about success. Verify everything. This is the **heavyweight, separate-session** tool. Its cheaper cousin — in-session subagents — covers most parallel work for free. **The Front Door decides which you need.** See `references/agent-army-vs-subagents.md` for the full comparison and `DECISIONS.md` for why the design is shaped this way. ## 🧭 The workflow: spec in one session, implement in another Agent-army is the **back half** of a two-session pipeline: 1. **Session A — spec.** Decompose the app into independent seams and write each seam's brief. The **spec source is pluggable**: an **OpenSpec** change dir, **Todoist** tasks, **JIRA** tickets, or just a **markdown brief** (even a near-empty stub the agent fleshes out). One seam = one spec = one