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 instructions/dheerg/swarms/agents-mdgit clone --depth 1 https://github.com/DheerG/swarmsWhat 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 | $0.04404 | $0.04404 |
| Opus 5 | $0.02202 | $0.02202 |
| Sonnet 5 | $0.00881 | $0.00881 |
| Haiku 4.5 | $0.00440 | $0.00440 |
Grade A, and why
swarms AGENTS.md 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 2d 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 — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Anti-ratchet constraint on the briefing templates
- The briefing templates are FIXED (canonical in skills/workflow-rules/SKILL.md). No added sections, no prescribed investigation steps, no "first action" items, no acknowledgment rituals. If a member needs more context, improve the noun-phrase identity in suggest-members — never the template. (Observed regression vector: f7db555 sprayed framing into every brief → premature-termination and step-repetition failures.)
- Carve-out: harness protocol mechanics are permitted. One instruction on HOW members communicate (SendMessage is the wire; plain text dies with the turn) is transport, not task prescription. Task framing, lifecycle framing, phase framing, and rituals remain forbidden.
What This Is
Swarm is a Claude Code plugin for launching agent teams. Eight commands drive an interactive setup that creates a coordinated team with defined roles and rules; users extend it with custom mode skills — full modes or thin wrappers over a built-in mode.
Architecture
Everything is a prompt. No runtime composition, no imports, no framework — self-contained markdown consumed in one pass.
commands/launch.md # Catch-all command — interactive team setup (Steps 0–8)
commands/code.md # Mode shortcut — pre-selects Code, delegates to launch.md
commands/triage.md # Mode shortcut — pre-selects Triage, delegates to launch.md
commands/write.md # Mode shortcut — pre-selects Writing, delegates to launch.md
commands/refine.md # Standalone — runs Review/Refine/Deliver against the current branch + PR
commands/workflow.md # Custom mode entry point — takes a mode skill name, delegates to launch.md
commands/create-workflow.md # Scaffolding — interviews user, generates mode skill + shortcut command (wrapper or full)
commands/update-workflow.md # Refresh — regenerates the plugin-owned wiring of an existing shortcut command
skills/code-mode/ # Code mode: lead identity, facilitator title, rules, phase arc
skills/triage-mode/ # Triage mode: diagnose an issue (cause + blast radius), no code change; phase arc has no Refine
skills/writing-mode/ # Writing mode: lead identity, facilitator title, ownership boundaries, editorial baseline, phase arc
skills/general-mode/ # General mode: silent fallback + wrapper base — no shortcut command
skills/workflow-rules/ # CANONICAL governance spec — hard rules, briefing templates, gate presentation contract, launch mechanics, pulse
skills/gate-presentation/ # Frozen per-gate constants (question/options/digest/preview) — invoked fresh at each gate
skills/suggest-members/ # Recommends team composition based on outcomes and mode
skills/writing-style/ # Structural pattern analysis (trope detection) for writing-mode review
skills/resolve-dispute/ # Resolves stuck review findings via put-up-or-concede exchange
skills/define-rubric/ # Available skill for teams that genuinely need formal validation criteria
skills/independent-review-loop/ # Independent pre-delivery review loop — Codex or swarm-native fallback
agents/swarm-member.md # Read-only team-member agent definition — every spawned teammate
agents/swarm-reviewer.md # Ephemeral read-only reviewer for the independent-review-loop fallback — never a teammate
.claude-plugin/plugin.json # Plugin manifest
.claude-plugin/marketplace.json # Marketplace registry entry
.claude/swarm-ship.md # Per-project ship definition (created at first launch, user-owned)
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.
- 2d ago First seen · 106 lines · 4,404 tokens per session scan A 8e2841368780
swarms AGENTS.md is an instructions file published in the GitHub repository DheerG/swarms (86 stars, last pushed 1mo ago), licensed MIT. It adds 4,404 tokens to every session, about $0.0220 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 instructions, from other repositories
moo.md CLAUDE.md
Instructions for saadshahd/moo.md, covering claude.md, what this is, conventions, philosophy and model-judgment boundaries.
pocketto AGENTS.md
Instructions for rfxlamia/pocketto, a project described as: A middleweight workflow for serious AI-coded features: less ceremony than BMAD, more accountability than Superpowers. Source : Trust Me Bro.
claude-sdlc CLAUDE.md
Instructions for lantisprime/claude-sdlc, covering claude.md, what this repo is, design intent — read this before "improving" anything, eat your own dog food and hook strictness philosophy.
kage-bunshin CLAUDE.md
Claude Code instructions for qkitzero/kage-bunshin, covering kage bunshin - second brain agent system, architecture, agents (.claude/agents/), skills (.claude/skills/) and notebook.
wormhole copilot-instructions.md
Copilot instructions for fatmali/wormhole, covering github copilot instructions for wormhole projects, 🚨 critical: wormhole logging protocol, mandatory workflow, 1. start session (required first step) and 2. check recent activity (before starting).
toh-framework CLAUDE.md
Instructions for wasintoh/toh-framework, covering claude.md — toh framework (repo development guide), what this is, everyday commands, verification protocol and single source, transformed per ide.