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 skills/callstackincubator/agent-skills/agent-devicenpx skills add callstackincubator/agent-skills --skill agent-devicegit clone --depth 1 https://github.com/callstackincubator/agent-skillsWhat 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 | $0.00053 | $0.01046 |
| Opus 5 | $0.00026 | $0.00523 |
| Sonnet 5 | $0.00011 | $0.00209 |
| Haiku 4.5 | $0.00005 | $0.00105 |
Grade A, and why
agent-device 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 — 65 lines — stays where its author put it; the contents beside it link to each section on GitHub.
agent-device
Use this skill as a router with mandatory defaults. Read this file first. For normal device tasks, always load references/bootstrap-install.md and references/exploration.md before acting. Use bootstrap to confirm or establish deterministic setup. Use exploration for UI inspection, interaction, and verification once the app session is open.
Default operating rules
- Start conservative. Prefer read-only inspection before mutating the UI.
- Use plain
snapshotwhen the task is to verify what text or structure is currently visible on screen. - Use
snapshot -ionly when you need interactive refs such as@e3for a requested action or targeted query. - Avoid speculative mutations. You may take the smallest reversible UI action needed to unblock inspection or complete the requested task, such as dismissing a popup, closing an alert, or clearing an unintended surface.
- Do not browse the web or use external sources unless the user explicitly asks.
- Re-snapshot after meaningful UI changes instead of reusing stale refs.
- Prefer
@refor selector targeting over raw coordinates. - Ensure the correct target is pinned and an app session is open before interacting.
- Keep the loop short:
open-> inspect/act -> verify if needed ->close.
Default flow
- Load references/bootstrap-install.md and references/exploration.md before acting on a normal device task.
- Use bootstrap first to confirm or establish the correct target, app install, and open app session.
- Once the app session is open and stable, use exploration for inspection, interaction, and verification.
- Start with plain
snapshotif the goal is to read or verify what is visible. - Escalate to
snapshot -ionly if you need refs for interactive exploration or a requested action. - Use
get,is, orfindbefore mutating the UI when a read-only command can answer the question. - End by capturing proof if needed, then
close.
What ships with it
7 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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 · 65 lines · 53 tokens per session scan A 2a0dd294991e
agent-device is a skill published in the GitHub repository callstackincubator/agent-skills (1,633 stars, last pushed 25d ago), licensed MIT. It adds 53 tokens to every session and 1,046 once invoked, about $0.0003 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 skills, from other repositories
canvas
Display, present, and render HTML content on connected Otto nodes (Mac app, iOS, Android). Show on device, preview on mobile, push to screen, or navigate to a URL on any connected node. Use for games, visualizations, dashboards, interactive demos, and live-reloading development previews. Supports presenting, hiding…
ios-simulator-browser
Stream an explicit iOS Simulator through pinned serve-sim into the T3 Code in-app browser or another agent browser. Use on Apple Silicon macOS when the user should watch simulator verification live or when browser-visible simulator evidence is needed.
argent-device-interact
Interact with an iOS simulator, Android emulator, or Chromium (CDP) app using argent MCP tools. Use when tapping UI elements, performing gestures, scrolling/swiping, typing text, pressing hardware buttons, launching apps, opening URLs, taking screenshots, waiting for an element to appear or disappear, or checking…
argent-tv-interact
Control and inspect TV apps via argent — Apple TV (tvOS), Android TV (leanback), and Amazon Fire TV (Vega). Boot the target, read focus, navigate with the D-pad remote, type, screenshot, and on Vega debug the JS runtime (evaluate, console logs, network inspector). Use when a task targets a TV (runtimeKind "tv", or…
asc-app-create-ui
Create a new App Store Connect app record via browser automation. Use when there is no public API for app creation and you need an agent to drive the New App form.
android-ui-automation
Automate Android apps that have no public API or web version by driving the UI layer through the Accessibility Service. Handles: deep-link launch (open an app into a specific screen), intelligent node location (find a button or text by fuzzy match, not exact text), tap-and-verify (confirm an action had its effect)…