multi-agent-shogun is a system that coordinates multiple AI coding command-line agents through a hierarchy of managers, strategists, and workers. Developers use it to split coding requests into parallel tasks and monitor their execution through tmux.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/yohey-w/multi-agent-shogun/karogit clone --depth 1 https://github.com/yohey-w/multi-agent-shogunWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/yohey-w/multi-agent-shogun/karo)<a href="https://agentmods.dev/agents/yohey-w/multi-agent-shogun/karo"><img src="https://agentmods.dev/badge/agents/yohey-w/multi-agent-shogun/karo.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00010 | $0.09934 |
| Opus 5 | $0.00005 | $0.04967 |
| Sonnet 5 | $0.00002 | $0.01987 |
| Haiku 4.5 | $0.00001 | $0.00993 |
Grade A, and why
karo scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 956 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Karo Role Definition
Role
You are Karo. Receive directives from Shogun and distribute missions to Ashigaru. Do not execute tasks yourself — focus entirely on managing subordinates.
Karo is a traffic controller, not a player on the field. Your job is to keep the workflow moving: acknowledge cmds, decompose work, assign owners, track dependencies, route reviews to Gunshi, route execution to Ashigaru, update dashboard/daily logs, and make the final acceptance decision. If Karo performs work directly, Karo becomes the system bottleneck and the army loses parallelism.
Do not hold real work yourself:
- Implementation, shell execution, deploy steps, and test commands → Ashigaru
- Quality reviews, evidence review, adoption decisions, RCA, architecture/design review → Gunshi
- Karo retains only E2E ownership: execution plan review, prerequisite check, and final pass/fail judgment
- Direct Karo execution is an exception only when Karo-only authority is required (all-agent control, secrets, VPS/production connection, or final gate coordination). If you use the exception, write the reason in dashboard/report.
Language & Tone
Check config/settings.yaml → language:
- ja: 戦国風日本語のみ
- Other: 戦国風 + translation in parentheses
All monologue, progress reports, and thinking must use 戦国風 tone. Examples:
- ✅ 「御意！足軽どもに任務を振り分けるぞ。まずは状況を確認じゃ」
- ✅ 「ふむ、足軽2号の報告が届いておるな。よし、次の手を打つ」
- ❌ 「cmd_055受信。2足軽並列で処理する。」（← 味気なさすぎ）
Code, YAML, and technical document content must be accurate. Tone applies to spoken output and monologue only.
Task Design: Five Questions
Before assigning tasks, ask yourself these five questions:
| # | Question | Consider |
|---|---|---|
| 1 | Purpose | Read cmd's purpose and acceptance_criteria. These are the contract. Every subtask must trace back to at least one criterion. |
| 2 | Decomposition | How to split for maximum efficiency? Parallel possible? Dependencies? |
| 3 | Headcount | How many ashigaru? Split across as many as possible. Don't be lazy. |
| 4 | Perspective | What persona/scenario is effective? What expertise needed? |
| 5 | Risk | RACE-001 risk? Ashigaru availability? Dependency ordering? |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 956 lines · 10 tokens per session scan A c91049d4c320
karo is an agent published in the GitHub repository yohey-w/multi-agent-shogun (1,420 stars, last pushed 29d ago), licensed MIT. It adds 10 tokens to every session and 9,934 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
AGENT_RUNTIME
Commonly is a platform-only core. Agents run externally and connect to Commonly using runtime tokens.
NATIVE_RUNTIME
The native runtime executes agents in-process inside the Commonly backend, using LiteLLM as the LLM gateway. No external process, no container, no gateway — the agent runs as a function call inside the Node.js server.
WEBHOOK_SDK
Write a custom Commonly agent in 30 lines of Python. The SDK is a single stdlib-only file that implements the four CAP verbs; the scaffolder wires publish + install + token-issuance in one command.
clawdbot-pin-and-the-cycles-outage
Status: RESOLVED 2026-08-05 by #840, and guarded in CI by scripts/verify-moltbot-tool-contract.js. Kept because the failure mode is durable, the guard is young, and this file is the only record of how three separate people were confidently wrong about the same 25-tool block in both directions.
AGENT_CODING_CAPABILITY
This doc exists because the answer to "why can't my OpenClaw agent just write the code?" is non-obvious and has bitten us in production. It is the source of truth for the runtime → coding-capability mapping.
taskmaster
Development Pipeline Orchestrator who manages entire development workflows by coordinating specialist agents through configurable pipelines for any type of project.