Trailblaze CLI
Natural-language CLI for iOS, Android and web screens. A saved session becomes a trail that CI replays.
- 332 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The agent drives a real device and can tap any element, including irreversible ones.
Install
- Works in
Manual install
brew install block/tap/trailblazeThe first install command from the README. The agent skill is added with `trailblaze skill install`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent drives a real device and can tap any element, including irreversible ones.
This is third-party code. Review the repository files before installing.
What it does
Trailblaze gives an agent the `trailblaze snapshot` and `trailblaze tool tap` commands: the agent reads the screen element tree and acts on a real device, emulator or browser. A good session is saved with `trailblaze session save` into a `trail.yaml` file. The file holds natural-language steps and recorded actions. `trailblaze run` replays them without calling a model, so CI runs are deterministic. The agent skill ships inside the CLI: `trailblaze skill install` puts it into `.claude/skills/` or `.agents/skills/`.
Who it is for. Mobile and web QA engineers who want to author scenarios by talking to an agent and then run them in CI without a model.
Good fit when
- A new sign-in flow is first walked through with an agent on an emulator.
- A good session becomes a regression test for CI.
- The same scenario is checked on Android, iOS and web.
- The team reads test steps as plain text.
Not a fit when
- You need a ready test framework with no CLI or agent.
- No device or emulator is available.
- The scenario only tests a server API.
Example request
Connect to the Android emulator, open Clock, set an alarm for 7:00 and save the session as a trail.Limitations
iOS needs macOS; Windows supports Android and web. Installing from the release script needs Java 17 or newer. Recording and self-healing depend on a model, while replay runs without one. Report videos need `ffmpeg`. README commands change device state, so run tests on a dedicated emulator or test account.
How to disable. Remove the skill from `.claude/skills/trailblaze/` or `.agents/skills/trailblaze/` and uninstall the CLI through Homebrew.
FAQ
Which platforms are supported?
Android, iOS and web. iOS runs only on macOS; Windows covers Android and web.
Is a model needed on replay?
No. `trailblaze run` replays recorded actions without calling a model.
How do I give the skill to an agent?
`trailblaze skill install` writes it to `.claude/skills/`; with `--agent agents` it goes to `.agents/skills/` for Codex, Cursor and others.
How do I check the skill is current?
`trailblaze skill status` compares the installed copy with the CLI version.
What is the license?
Apache-2.0.
Are there other install options?
Yes, `install.sh` for macOS and Linux and `install.ps1` for Windows; Homebrew is listed first.
Discussion
No questions or reviews yet.
Related
CLI toolsRuns Maestro flows on Android, iOS and in the browser as a single binary without a JVM and installs a skill for the agent.
The noqa-testing skill teaches an agent to test iOS and Android: read the screen, tap, write cases and run them on a real phone
The Revyl platform CLI: an agent runs a mobile app on a cloud device and checks it with tests written in plain language
Microsoft’s CLI controls real terminal sessions, checks visible text and records test runs for investigating failures.
CLI, MCP server and typed Node.js API so an agent can verify its changes in a live mobile app on iOS, Android and more
A skill that teaches Claude Code and Cursor to wire in Tracely tracing and turn failing agent runs into CI tests
README in short
A session with an agent is saved as a single trail.yaml file. People read the steps, and CI executes the recorded actions without a model.