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/nmamano/isomux/agents-mdgit clone --depth 1 https://github.com/nmamano/isomuxWhat 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 | $0.01949 | $0.01949 |
| Opus 5 | $0.00975 | $0.00975 |
| Sonnet 5 | $0.00390 | $0.00390 |
| Haiku 4.5 | $0.00195 | $0.00195 |
Grade A, and why
isomux 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 — 65 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Isomux
Isomux is a meta-harness: it sits one level above Claude Code, Codex, and OpenCode and manages multiple agents, adding inter-agent messaging, a shared task board, human collaboration, a mobile UI, and more. See the README.md for a full feature overview and setup instructions.
How to develop
The running server is on localhost:4000, managed by systemd as the user-level service isomux.
Server-side code changes always require a restart. Restarting the server is mildly disruptive: it stops every agent in the office until the user re-engages them.
For UI changes, run bun run build:ui. Outputs to ui/dist/, which the server reads per-request. No server restart needed, only a page refresh.
On a fresh checkout or worktree, run bun run build:ui before bun test. Tests that serve the app shell or PWA assets need the built files.
If a run reports a test timeout, run the affected file alone before you attribute later failures in that file. As measured 2026-08-29, Bun 1.3.11, 1.3.14, and 1.4.0 do not stop a timed-out test body. It can continue to change shared state after teardown and make later tests fail.
Proactively remind the user of what steps they need to take to do what they want: e.g., to test a front-end only change, or a server change in main, or a change in a worktree. Offer to do the steps yourself. After completing a feature or batch of fixes, offer to the user to commit.
If an action runs into permission issues because it's a destructive action, let the user know and give them the command they need to run to do it themselves.
Debug agent issues by reading logs at ~/.isomux/logs/<agentId>/<sessionId>.jsonl.
Look at UI changes instead of assuming them. A box set up by deploy/install.sh normally has a headless Google Chrome at /usr/bin/google-chrome (the installer warns and carries on when it could not install one), and there are two ways to use it: the preview-url card, which screenshots a page straight into the chat, and Playwright launched with channel: "chrome", which drives that same Chrome. Taking that route, Playwright needs no browser download of its own: no npx playwright install, and for Chromium no playwright install-deps either, since Chrome's package already pulls in the shared libraries it would install.
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 · 65 lines · 1,949 tokens per session scan A 0f0f6b603171
isomux AGENTS.md is an instructions file published in the GitHub repository nmamano/isomux (37 stars, last pushed 2d ago), licensed MIT. It adds 1,949 tokens to every session, about $0.0097 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
babysitter AGENTS.md
Instructions for a5c-ai/babysitter, covering agents.md, repo-specific overrides, reference map and quick commands.
open-bridge AGENTS.md
Instructions for bks-lab/open-bridge, covering the bridge — agent instructions, required reading, session start detection (automatic), theme and agents.
agents CLAUDE.md
Instructions for eloylp/agents, covering claude.md, project overview, directory structure, config model and build & run.
designpowers GEMINI.md
Gemini CLI instructions for Owl-Listener/designpowers, covering designpowers, mandatory: welcome sequence first, skills, agents and notes for gemini cli.
kandev CLAUDE.md
Claude Code instructions for kdlbs/kandev, a project described as: AI Kanban & Development Environment. Orchestrate multiple agents, review changes, open PRs. Multi-provider, self-hostable, no telemetry.
psyche-build AGENTS.md
Instructions for OpenCoven/psyche-build, covering psyche build agent guide, repository role, planning and evidence contract, canonical routing and deterministic entrypoints.