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/multiplex-term/multiplex/e2e-headlessgit 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 | $0.00000 | $0.05300 |
| Opus 5 | $0.00000 | $0.02650 |
| Sonnet 5 | $0.00000 | $0.01060 |
| Haiku 4.5 | $0.00000 | $0.00530 |
Grade A, and why
e2e-headless 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 — 311 lines — stays where its author put it; the contents beside it link to each section on GitHub.
End-to-end verification (headless, no real remote needed)
Tools/dev-sshd/harness.sh runs a user-mode sshd on 127.0.0.1:2222 (own
keys under Tools/dev-sshd/state/, never touches ~/.ssh):
./Tools/dev-sshd/harness.sh start # keys + sshd + writes state/seed.json
./Tools/dev-sshd/harness.sh demo # tmux sessions: main, scratch, deploy, agent
./Tools/dev-sshd/harness.sh stop
The agent session fakes CLI agents: window cc sets the ✳ Claude Code
OSC title, cx runs exec -a codex cat, pi sets π - harness + exec -a pi cat; all run cat, so tmux capture-pane -t agent:0 -p shows what a
chip typed.
DEBUG-only env vars (the simulator shares the Mac's network; pass with
the SIMCTL_CHILD_ prefix through xcrun simctl launch <UDID> app.multiplexterm.multiplex):
MULTIPLEX_SEED_HOST=<path to seed.json>— imports a readydevboxhost (idempotent, stable UUID). ⚠ Idempotence is per install:simctl uninstallwipes hosts.json but NOT the Keychain mirror, so the next launch seeds a new devbox AND re-adopts the old record (two identically named hosts). Prefer terminate + relaunch over reinstall. Optional seed keys:workingDirs/sessionScripts/newSessionTmuxConf/agentLaunchModelsfeed the picker/script/conf/model paths;secondaryBackends: ["herdr"]imports a MIXED host (two probes on one connection, both backends' tiles);passphrasestores the key passphrase in the Keychain so an encryptedprivateKeyproves the sealed-key connect path (probe logsAccepted publickey, no prompt);"enabled": falseproves the never-dials promise against a silentstate/sshd.log;"backgroundKeepAlive": truestarts the host opted into background keep-alive (nothing can tap that switch headlessly).MULTIPLEX_AUTO_ATTACH=main,scratch— opens a terminal per entry via the real Attach route;+inside an entry groups sessions as tabs of one window. Fires once per process.MULTIPLEX_AUTO_ATTACH_HOST=devboxnames the host — without it the first host wins (on a device with synced real hosts that is NOT the seed).MULTIPLEX_AUTO_TMUX_COPY=1— sends Copy Mode after attach; proof:#{pane_in_mode}becomes 1.MULTIPLEX_AUTO_TMUX_CLOSE=pane|window— runs the confirmed close action through the direct SSH control path. Disposable sessions only; verify host-side.MULTIPLEX_AUTO_MERGE=1— merges every terminal window into the first through the real surrender/adopt path (moved tabs keep connections).MULTIPLEX_AUTO_DROP=<local path>— drops that file into the first tab (SFTP upload + typed path, the real drag path).MULTIPLEX_AUTO_ACTION_URL=multiplex://open?…— submits the URL through the exactonOpenURL→ router seam once per process; the headless widget stand-in (simctl openurl's first-run confirmation can't be clicked headlessly, and idb HID taps died with Xcode 27).MULTIPLEX_AUTO_BIND=<multiplex://b/…>/MULTIPLEX_BIND_AUTOPIN=<6 digits>— drive Bind Host headlessly: submit a payload through the real parse → confirm → handshake seam / answer the first machine heard with that PIN. Stage the other side withharness.sh bind(prints the PIN) orbind --print-only(prints the payload URL; transcript instate/bind.log). Proof is host-side: amultiplex:bind:<id>:<device>line instate/authorized_keys, thenAccepted publickeyinsshd.log.MULTIPLEX_BIND_PASSPHRASE=<text>presets KEY PASSPHRASE so a headless bind stores its key sealed;MULTIPLEX_BIND_BACKEND=<comma list>presets the pane's Backend selection, so the minted host record carries it (proof: the deck tile probes herdr, not tmux). The FIRST entry is the default that mints, and every entry is shown —herdris the old single-backend meaning unchanged,tmux,herdrbinds a MIXED host straight from the pane. ⚠ These areDeckWindowtasks — a restored terminal-only scene never runs them;simctl uninstall+simctl keychain <udid> resetfirst (or the mirror re-adopts the old host and the free host limit blocks the bind).MULTIPLEX_AUTO_ADD_HOST=bind|manual— opens Add Host on either road for layout capture (deliberately bypasses the free-tier add gate).MULTIPLEX_PRO_LOCKED=1— free-tier mode, never persisted; withMULTIPLEX_AUTO_ATTACH=agent+ thedebug.agentchiphook it proves exactly ten slash-chip sends.MULTIPLEX_DECK_SIZE=<w>x<h>/MULTIPLEX_TERM_SIZE=<w>x<h>(visionOS) — scene default window sizes for screenshot runs (the sim can't drag-resize).MULTIPLEX_AUTO_PAYWALL=1— the real locked paywall with a deterministic $19.99 storefront preview (simctl launches don't inherit Xcode's StoreKit session).MULTIPLEX_AUTO_SETTINGS=1|theme|licenses/MULTIPLEX_AUTO_FAQ=1— open those sheets for headless capture (themepushes the theme editor,licensesthe Open Source Licenses page).MULTIPLEX_AUTO_WHATS_NEW=1|log— open the release-notes launch card / the full record behind its FULL NOTES chip. The launch gate itself is unreachable headlessly (it fires only for an install carrying a PREVIOUS version's defaults), so this is the layout road; the gate is pinned byDeckWindowUIKitTestsinstead. ⚠ Both screens areDeckWindowpresentations — a restored terminal-only scene shows nothing — and the notification-permission alert iOS raises on a fresh install sits over the middle of the sheet with no headless way to dismiss it.MULTIPLEX_AUTO_HOST_SETTINGS=1|models|backend|directories— opens the first host's edit sheet (regression-checks the Observation environment across the sheet boundary — a missing HostStore is a fatal error);modelsscrolls to the Agent launch models section,backendto the tmux/herdr Backend section,directoriesto the New session defaults working-directories editor (shown for BOTH backends — the herdr mint roots a session's world with these; only the tmux options editor stays tmux-scoped).- Attention alerts need Pro AND the
attention.alertsEnabledswitch; both live in the app container's defaults, andsimctl spawn <UDID> defaults write <bundle-id>lands in a domain the sandboxed app never reads — pass the container plist path as the domain instead (… defaults write <container>/Library/Preferences/app.multiplexterm. multiplex.plist MultiplexProUnlocked -bool true, same forattention.alertsEnabled; container viasimctl get_app_container … data). A stalefalsethere logsalert dropped … locked(categoryattention) exactly like the Pro gate. Proof of a live edge:log stream --level debug --predicate 'category == "attention"'showsalert posted for <session>on the probe tick after the agent's title drops its spinner (Grok Build, 2026-08-16: asleep 20turn against this Mac's tmux). MULTIPLEX_KEYCHAIN_TIP=locked|unlocked|missing— forces the keychain verdict. The sign-in-screen gate still applies: inject a needle first (e.g.tmux send-keys -t agent:cc 'Select login method:' Enter).MULTIPLEX_APP_LOCK=1|held— start behind the app-lock veil (never persisted).1keeps the real authenticator;heldrefuses every attempt so the veil can be captured. Shipping toggle: Settings → App lock (AppLockStore).MULTIPLEX_FORCE_SHELL=1|0— force the single-window shell on/off. Default: iPhone always shell, iPad only whenUIWindowScene.isFullScreen, visionOS never. Logged under categoryshell.MULTIPLEX_SIDE_PANEL=0— DEBUG-only legacy road: terminal-confirmed ▤/⌗ viewers dock as tabs instead of side panels. PassSIMCTL_CHILD_MULTIPLEX_SIDE_PANEL=0to preserve older recipes.
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 · 311 lines · 0 tokens per session scan A eac4f814e085
e2e-headless is an agent published in the GitHub repository multiplex-term/Multiplex (10 stars, last pushed 5d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 5,300 tokens. 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 agents, from other repositories
layer3-issue-detection
Layer 3 systematically scans ALL entry points from Layer 1 and applies issue detection rules. Unlike Layer 2 (which traces specific flows in depth), Layer 3 does a breadth-first scan to categorize issues across the entire codebase.
layer4-semantic-evaluation
Layer 4 evaluates workflows from the user's perspective. While Layers 1-3 analyze code structure, Layer 4 asks: "Does this workflow help the user achieve their goal?".
layer2-methodology
Layer 2 takes entry points from Layer 1 and traces the complete user journey.
mobile-architect
Mobile architecture expert. Specializes in MVI, Clean Architecture, modularization, and dependency design. Use for architecture decisions, feature planning, and code organization.
Native SwiftUI Builder
Builds complex custom SwiftUI views using standard SwiftUI components with TTBaseUIKit design tokens (XView/XSize/XFont). Does NOT use TTBaseSUI wrapper components.
kmp-architect
KMP architecture expert. Designs shared module structures, expect/actual patterns, navigation, and DI for Kotlin Multiplatform projects.