OpenCLI Usage Guide
Turn any website into a CLI and drive it with your logged-in browser via AI agents.
The skill states credentials are extracted from the live Chrome session and stored in a cache, but does not disclose data flow, storage location, or user confirmation mechanisms. Permissions are limited to Bash(opencli:*) and Read, which is conservative, but no rollback or isolation mechanisms are provided. Deductions: insufficient transparency on sensitive data handling and data flow.
Documentation is detailed, and CI provides build and unit tests, but there is no direct test evidence for the skill's key paths (e.g., browser binding, adapter discovery). Real-site regression tests skip transient failures. Deductions: static review cannot verify key paths, and test coverage is incomplete.
Clearly defines suitable scenarios (e.g., adapter discovery, browser driving) and non-suitable ones (e.g., depending on browser without extension installed), and provides Chinese-language support (e.g., README zh-CN). Deduction: core functionality depends on Chrome extension from Chrome Web Store, which may be unreachable from mainland-China networks.
Well-structured documentation with progressive disclosure, installation notes, examples, environment variables, and known limitations (e.g., removed commands). Deduction: lacked version history and maintenance responsibility statement.
Skill describes core tasks (e.g., discovering adapters, driving browsers) but does not provide directly usable output examples or verification results. Deduction: static review cannot verify direct usability of outputs.
CI workflows and test suites exist, but they do not cover the skill's key paths, and this is a static review. Deduction: lack of third-party execution evidence.
- The skill claims to extract credentials from the Chrome session, but does not disclose storage location or cleanup mechanisms, posing a risk of sensitive information exposure.
- Core functionality depends on a Chrome extension from the Chrome Web Store, which may be unreachable from mainland-China networks, or requires manual installation.
- User confirmation mechanisms are unclear: the skill may perform browser actions (e.g., clicks, form fills) on behalf of the user without explicit confirmation.
- The skill references external commands (e.g., yt-dlp) without providing installation or version requirements.
What does this skill do, and when should you use it?
OpenCLI transforms websites, Electron apps, and external CLIs into a unified `opencli <site> <command>` interface for agents. This skill is the orientation layer for an OpenCLI session: it maps the three pillars (adapter commands, browser driving, external CLI passthrough), explains how to discover adapters, lists universal flags and output formats, and routes to the next specialized skill. Agents should always query `opencli list -f json` for live command lists instead of relying on static documentation.
Executes opencli list -f json to get the current command registry; uses opencli <site> --help to inspect site-specific commands; supports -f flags (json, plain, yaml, table, md, csv) and -v for verbose logging; runs opencli doctor to diagnose the browser bridge; reads environment variables for timeouts and CDP overrides.
- An AI agent starting an OpenCLI session loads this skill to understand the overall capability map.
- An agent asked 'what can opencli do?' runs `opencli list -f json` to get the live command list.
- An agent needs to find adapters for a specific site, filtering `opencli list -f json` by keyword.
- Before browser-dependent work, an agent runs `opencli doctor` to verify bridge connectivity.
- An agent deciding which specialized skill to load (e.g., opencli-browser, opencli-adapter-author) consults this skill's routing table.
What are this skill's strengths and limitations?
- Real-time command discovery avoids doc rot
- Clearly specifies strategy-dependent prerequisites (PUBLIC/COOKIE/INTERCEPT/UI/LOCAL)
- Provides self-repair workflow with `--trace retain-on-failure`
- Multiple output formats for agent parsing (json, yaml, plain, etc.)
- Requires Node.js >= 21; browser-dependent features need extension and logged-in sessions
- Browser commands require `opencli doctor` to pass, adding environmental complexity
- No explicit test suite or coverage evidence; reliability rests on autofix
- Electron app adapters require the app to be running
How do you install this skill?
Install the collection: npx skills add jackwener/opencli --skill opencli-usage, or copy the skills/opencli-usage folder into your agent's skills directory.
How do you use this skill?
Trigger in agent prompts: when a task involves OpenCLI, first call opencli list -f json to discover commands. For example, opencli hackernews top --limit 5. Follow the skill's guidance to choose between adapters and raw browser commands.
How does this skill compare with similar options?
Compared to raw browser automation tools like Playwright or Puppeteer, OpenCLI offers higher-level adapter abstraction and session reuse, but requires installing the opencli CLI and a browser extension.