Use with AI Agents

Qirabot can be operated by an AI agent as well as called from code. A pre-built skill following the Agent Skills open standard equips the agent with a preflight environment check, condensed SDK and CLI references, and per-platform starter templates. Given a natural-language automation goal, the agent validates the environment, selects the execution path (the CLI for one-shot tasks, an SDK script when the flow requires branching or returned values), and verifies the run's outcome. Installation depends on the agent.

Claude Code plugin

The qirabot plugin packages the skill for Claude Code's plugin marketplace:

/plugin marketplace add qirabot/claude-plugins
/plugin install qirabot@qirabot

Claude invokes the skill automatically whenever a task involves automating, testing, or scraping a UI; it can also be invoked explicitly as /qirabot:qirabot. The skill contains:

• A preflight script validates the Python environment, backend dependencies, and Google Cloud credentials (ADC) before any code is written, and prints the exact fix command for each failing check.
• Condensed SDK + CLI references: the agent codes against an accurate, drift-tested API surface.
• Starter templates for browser, Android (adb), iOS (WDA and Appium), and bring-your-own-driver integration, so the agent adapts a working skeleton instead of generating boilerplate from scratch.

The plugin contains instructions and helper scripts only; the qirabot package itself is installed at runtime by the preflight bootstrap. The marketplace copy auto-updates with the repository's main branch; for a copy pinned to the installed SDK version, use qirabot skill install claude instead.

Any other agent (Codex, Cursor, Copilot, …)

Installing the skill. The Agent Skills format is supported by Codex, Cursor, Gemini CLI, and many other tools. The same skill the Claude Code plugin ships is bundled in the qirabot package, so install qirabot first:

qirabot skill install agents            # the shared .agents/skills convention
qirabot skill install codex             # or: claude, cursor
qirabot skill install --dir <path>      # any other Agent-Skills-compatible tool

The installed copy is version-matched to the SDK: the API reference the agent reads always describes the qirabot it runs. --project installs into the repository (.agents/skills/) instead of the home directory; rerun the command after upgrading qirabot. For details, see the CLI Reference.

Alternatively, the skills CLI installs the same skill from the repository's main branch (latest instructions, not pinned to the installed SDK version):

npx skills add qirabot/qirabot

Without the skill

Independently of the skill, two properties make Qirabot straightforward for any agent to operate:

The CLI functions as an agent tool. A single shell command executes a complete natural-language task; one-shot jobs require no code generation:

qirabot browser "Fill the signup form as Jane Doe and stop at the captcha" --url example.com

Exit codes are machine-checkable (0 pass, 1 fail, 130 interrupted), and every run writes an HTML report with per-step screenshots for the agent or a human to inspect on failure. All commands are listed in the CLI Reference.

The documentation is agent-readable. Reference these in the agent's context or rules file:

• https://qirabot.com/docs/llms.txt: an index with per-page summaries
• https://qirabot.com/docs/llms-full.txt: the complete documentation in one file
• every page as raw Markdown by substituting .md for .html, e.g. https://qirabot.com/docs/reference/methods.md

In Cursor, add them via the @Docs feature.

Why agents and vision automation fit

An agent generating Playwright code must guess selectors it cannot observe, and those selectors break on the next markup change. With Qirabot the agent addresses elements in natural language, the modality it reasons in, and the same skill extends to surfaces that code-first stacks cannot reach: native mobile apps, desktop software, and games. See What is Qirabot and the platform support matrix.