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/etanhey/cmuxlayer/agents-mdgit clone --depth 1 https://github.com/EtanHey/cmuxlayerWhat 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 | $0.00559 | $0.00559 |
| Opus 5 | $0.00280 | $0.00280 |
| Sonnet 5 | $0.00112 | $0.00112 |
| Haiku 4.5 | $0.00056 | $0.00056 |
Grade A, and why
cmuxlayer 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 3d 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 — 51 lines — stays where its author put it; the contents beside it link to each section on GitHub.
cmuxlayer
What this repo is for
cmuxlayer should be a harness for harnesses — one thing that can control every other harness with a single truth.
Codex, Kiro, Cursor CLI, whatever's next. I'm not locked into one and I don't want to be. And they don't die when they're done: formal agents that persist, controlled by one agent.
The panes are already visible — I can go look myself. The point is that orc, the leads, and the other workers know what each other are doing, which model they're on, how many tokens they've burned and how much of their context window is left, without anyone opening a pane. And when a pane dies and something reboots it to resume, it's the same agent, on the right model, and cmuxlayer knows how to change that model on whichever harness it happens to be.
That's the single truth. Everything else here is downstream of it.
The one principle
Where a pane lands is not something an agent should have to remember. It should be deterministic enough that they can't really get it wrong — that's the point of this repo.
A workspace is my mental model of what's scoped. Leads left, workers right, same workspace. Ninety-nine percent of the time I won't ask for anything else, so anything that can quietly put it somewhere else is broken.
Resumable, automatically
Any lead or orchestrator should be able to resume an agent by its ID from the registry — I shouldn't have to go find the session and run resume myself. Mostly that's one of two cases: a worker got killed because its pane broke, or a lead thought it was done and wants it back.
If the session is old and lives in the Drive-mounted folder, cmuxlayer should pull it by agent ID on its own. It should be automatic and work right.
The tools are the product
cmux owns the terminal socket — don't rebuild it. But cmuxlayer must always be able to replace or fix its own backend without killing the MCP connection.
And keep what the tools hand back concise and to the point. They give us way too many rows of data — it eats tokens for no reason.
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.
- 3d ago First seen · 51 lines · 559 tokens per session scan A a7d5f714a13a
cmuxlayer AGENTS.md is an instructions file published in the GitHub repository EtanHey/cmuxlayer (26 stars, last pushed 3d ago), licensed Apache-2.0. It adds 559 tokens to every session, about $0.0028 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
mcp-server-excel coverage-prevention-strategy.instructions.md
Instructions for sbroenne/mcp-server-excel, covering generated surface coverage, contract change workflow, required checks and common incomplete changes.
mcp-server-excel mcp-llm-guidance.instructions.md
Instructions for sbroenne/mcp-server-excel, a project described as: Automate real Microsoft Excel with AI via MCP Server or CLI — Power Query, DAX, VBA, PivotTables, charts, and 326 operations.
openrouter-mcp-multimodal AGENTS.md
Instructions for stabgan/openrouter-mcp-multimodal, covering agent instructions, before you ship, releasing (read this before publishing), short version and version files (must all match package.json).
ntfy-mcp-server AGENTS.md
Instructions for cyanheads/ntfy-mcp-server, covering developer protocol, what's next?, core rules, patterns and tool.
Plonk AGENTS.md
Instructions for ostapondo/Plonk, covering agent rules, layout, adding a module, build & verify and code style.
mcp-dockhand copilot-review.instructions.md
Instructions for strausmann/mcp-dockhand: When reviewing pull requests in this repository.