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/multiplex-term/multiplex/agents-mdgit clone --depth 1 https://github.com/multiplex-term/MultiplexWhat 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 | $0.01630 | $0.01630 |
| Opus 5 | $0.00815 | $0.00815 |
| Sonnet 5 | $0.00326 | $0.00326 |
| Haiku 4.5 | $0.00163 | $0.00163 |
Grade A, and why
Multiplex 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 yesterday.
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 — 122 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Multiplex — a spatial SSH terminal built around remote tmux (with herdr
as an optional second backend); visionOS-first, iPadOS alongside. The deck is
a fleet-wide monitor wall: every host probes concurrently, every session
renders as a live capture-pane tile, and each attach opens its own terminal
window (a scene on visionOS / Stage Manager on iPad). Product tour:
README.md; visual system + rationale: DESIGN.md (the TALLY identity;
bake-off record in docs/design-bakeoff.md).
Build & test
The Xcode project is generated by XcodeGen — edit project.yml, never the
.xcodeproj (git-ignored). After changing project.yml, adding/removing
sources, or touching a vendored package's manifest: xcodegen generate.
Build / test / run use Tools/build.sh (see the build-and-verify skill for
the full workflow). Quick reference:
# Build
xcodebuild -project Multiplex.xcodeproj -scheme Multiplex \
-destination 'platform=visionOS Simulator,name=Apple Vision Pro' \
-derivedDataPath DerivedData build
# Unit tests — fast, no device I/O
xcodebuild -project Multiplex.xcodeproj -scheme MultiplexTests \
-destination 'platform=visionOS Simulator,name=Apple Vision Pro' test
# iPad target: -destination 'platform=iOS Simulator,name=iPad Pro 13-inch (M5)'
Always keep -derivedDataPath DerivedData, and never run iPad and visionOS
builds concurrently against it (the build DB locks).
Lint: ./Tools/build.sh lint (brew install swiftlint; --fix applies
correctable rules). Runs --strict: the tree stays at zero violations.
Deliberately not an Xcode build phase (script sandboxing would demand a
regenerated input-file list, and contributors without SwiftLint must still
build).
.swiftlint.yml deviates from the defaults only with an in-file reason —
keep it that way (size metrics, trailing_comma, and
optional_data_string_conversion are off by design, reasons in the file).
Where one line genuinely can't comply, use a scoped
// swiftlint:disable:next <rule> with the reason above it; prefer fixing
the code.
CI is split by platform and filters out unrelated paths. The macOS workflow
(.github/workflows/ci-macos.yml, macos-26 — Xcode 26 and the visionOS 26
simulator runtime) is Tools/build.sh gen, lint, then build + test for
visionOS and iPad in sequence, sequential because of the DerivedData lock
above. The Linux workflow (.github/workflows/ci-linux.yml) runs only when its
metadata inputs change. It executes ruby Tools/check-metadata.rb, which needs
neither Xcode nor gems and so answers in seconds: it holds the TestFlight
changelog and the App Store listing files to the App Store Connect caps and
the vetted glyph set in Tools/metadata_limits.rb (shared with
fastlane/Fastfile, so a lane and a pull request can never disagree about
them). Run it locally before committing metadata.
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.
- yesterday First seen · 122 lines · 1,630 tokens per session scan A 0ab72c226985
Multiplex AGENTS.md is an instructions file published in the GitHub repository multiplex-term/Multiplex (10 stars, last pushed 5d ago), licensed Apache-2.0. It adds 1,630 tokens to every session, about $0.0081 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 2026-08-31.
Other instructions, from other repositories
psmux AGENTS.md
Instructions for psmux/psmux, covering agent instructions and test isolation.
MarkdownView AGENTS.md
Instructions for keitaoouchi/MarkdownView, covering agents, technical components, relationships (data/event flow overview), extension points (overview) and platform / distribution.
liney AGENTS.md
Instructions for everettjf/liney, covering liney repository collaboration guide, 交流语言 / communication language, project overview, repository layout and build and test.
starboard CLAUDE.md
Instructions for palamim/starboard, covering claude.md, commands, code style, architecture and non-obvious gotchas (not discoverable by reading the code).
CommandMate CLAUDE.md
Instructions for Kewton/CommandMate, covering claude.md, ⚠️ 重要: モジュール詳細を claude.md に書かないこと, プロジェクト概要, 基本情報 and 技術スタック.
CommandMate AGENTS.md
Instructions for Kewton/CommandMate, covering agents.md, scope, project context, working principles and coding rules.