codebase-memory-mcp
Resolve canonical Git checkouts, index and verify codebase-memory-mcp graphs through the guarded CLI, open its browser viewer on…
Prepare and check trusted macOS hosts for physical Android and iPhone tests with agent-device, ADB, and Xcode, or set up iPhone Mirroring separately.
$ npx -y skills add vincentkoc/dotskills --skill agent-device-lab --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/agent-device-labContext preview
The summary Claude sees to decide when to auto-load this skill.
Prepare and check trusted macOS hosts for physical Android and iPhone tests with agent-device, ADB, and Xcode, or set up iPhone Mirroring separately.
name: agent-device-lab description: Prepare and check trusted macOS hosts for physical Android and iPhone tests with agent-device, ADB, and Xcode, or set up iPhone Mirroring separately. license: MIT metadata: source: "https://github.com/vincentkoc/dotskills"
Prepare a Mac and one physical Android or iPhone for an authorized test. This skill owns host readiness and device enrollment. Use the installed upstream `agent-device` skill or version-matched CLI help for app-driving commands.
For ordinary app testing on an already prepared target, use the app automation workflow directly. Simulator setup and fleet deployment are separate tasks.
1. Identify the requested mode, host, physical device, app, and observable result. An audit or plan does not authorize installation, enrollment, settings changes, or app interaction. Continue within existing authorization; obtain only any missing authority for the specific change. 2. Inspect only the selected mode's prerequisites. Mirroring alone needs no ADB, XCTest runner, or agent-device installation. For agent-device, resolve the installed binary and version; follow its `help workflow` and current [installation requirements](https://oss.callstack.com/agent-device/docs/installation). Prefer a trusted installed or project-pinned CLI. Installing a skill does not install that CLI; do not silently fetch and execute a mutable `@latest` package. 3. Report missing prerequisites before changing them. Limit authorized host preparation to official platform tools and the selected device platform. Device Trust/RSA prompts, passcodes, Developer Mode confirmation, Apple Account, 2FA, and signing-account authentication belong to the owner. Wait at that boundary; never collect their credentials or bypass the prompt.
Use Google's [Android SDK Platform Tools](https://developer.android.com/tools/adb). The owner enables USB debugging and approves this host's RSA key on the phone. `adb devices -l` must identify the intended physical serial in state `device`; `unauthorized`, `offline`, or a missing device is a blocker. A `device` entry alone does not prove Android has finished booting. Bind subsequent ADB commands with `-s <serial>` and agent-device with `--platform android --serial <serial>`.
Discovery may start the local ADB server. Do not restart a shared server or reset device authorizations as an automatic repair. Wireless pairing and helper/IME installation require task scope beyond checking a USB connection.
Use full Xcode with support for the phone's OS. Check the selected developer directory and Xcode version before proposing a switch; license acceptance and first-launch setup are host changes. Do not add simulator or watchOS runtimes for a phone-only task.
The owner completes [device trust](https://support.apple.com/109054) and [Developer Mode](https://developer.apple.com/documentation/xcode/enabling-developer-mode-on-a-device), including its restart and confirmation. Verify the exact phone in `xcrun devicectl list devices`; use `--platform ios --udid <udid>` for agent-device. Keep the phone unlocked for discovery and XCTest preparation. XCTest interaction also needs working Developer Disk Image (DDI) services and runner signing. A paired device is insufficient proof. For a DDI failure, inspect Developer Mode, Xcode/OS compatibility, and the exact CoreDevice diagnostic; do not copy unverified disk images or delete shared developer state.
Check Apple's current [Mirroring requirements](https://support.apple.com/120421), including supported hardware/OS and region, the same Apple Account with 2FA, Wi-Fi/Bluetooth, and a nearby locked iPhone. Confirm the intended phone when several are available. Mirroring is independent of XCTest/DDI readiness.
For new setup, prefer **Ask Every Time** and decline mirrored notifications unless requested. Preserve an existing explicit user preference. The owner completes Mac authentication and any phone-side prompt. Do not inspect notifications or personal content to demonstrate success. An unavailable region or unresolved account requirement ends with a concrete blocker, not a workaround.
When a device interaction is authorized, use a task-owned session and one named test surface, preferably the user's test app. Select a non-personal Settings page only if that is the agreed proof. Use `umask 077` for new agent-device state and evidence; inspect the resolved state directory before reuse. Do not change permissions or remove locks in another session's state.
Perform the bounded interaction, verify the expected result, and close only the session or Mirroring window opened for this task. If setup or interaction fails, report the failed prerequisite and the owner's next action. Keep unfinished recovery evidence; do not turn closeout into daemon, device, or cache cleanup. Scrub device identifiers, account names, screenshots, and logs before sharing.
The canonical source is [callstack/agent-device](https://github.com/callstack/agent-device). If installing its skill is requested, the standalone installer works without a dotskills checkout. Run from the intended project and choose the intended agent and installation scope:
npx skills add callstack/agent-device --skill agent-device
This installs instructions, not the runtime. If the skill is absent, installed CLI help remains usable. Do not run repository-wide import or sync to prepare one device. See upstream [agent setup](https://oss.callstack.com/agent-device/docs/agent-setup).
We are moving from "prompt as text" to skill as runtime module. Each skill acts like a lightweight, containerized application for AI work: a stable interface, opinionated workflow, and bundled resources that can be installed, versioned, tested, and reused
Resolve canonical Git checkouts, index and verify codebase-memory-mcp graphs through the guarded CLI, open its browser viewer on…
Mine structured Codex /goal history locally or across a configured machine fleet, measure active goal time and resumed thread…
Recover Codex or Claude sessions from tmux panes, saved cockpit snapshots, session history, and recent agent logs without…
Interpret Crabpot and OpenClaw performance dashboard metrics, import-loop profiles, runtime profiles, and branch-to-branch perf…
Maintain and release the crawlkit Go library, preserving downstream compatibility for gitcrawl, slacrawl, discrawl, and notcrawl.
Template skill for repository authors; excluded from public publishing.