jackwener/opencli
10 skills65.4k installs238k starsGitHub
Install
npx skills add https://github.com/jackwener/opencliSkills in this repo
1Opencli Usageopencli-usage is the orientation skill for OpenCLI: it explains how websites, Electron apps, and external CLIs collapse into a single `opencli` command tree that coding agents can execute instead of brittle screen automation. Solo builders reach for it at the start of a session when they or the agent ask what OpenCLI can do, how to find the right adapter under `clis/` or `~/.opencli/clis/`, and which follow-on skill covers browser driving or passthrough. The skill frames three pillars—adapter commands, browser driving, and tab binding—so you load specialized docs only after the mental model is clear. It is procedural meta-knowledge rather than one fixed integration, which is why it stays useful across build, ship, and operate work whenever you automate a vendor console or internal tool.13.4kinstalls2Opencli Browseropencli-browser teaches coding agents how to operate opencli’s browser subcommands against a real Chrome window: run doctor first, pick a stable session name, then drive inspect, click, and form flows using selector-first targets and the structured envelopes the CLI returns. It is aimed at solo and indie builders who need ad-hoc automation—checking a dashboard, walking through OAuth or logged-in UI, or scraping state—without maintaining a full Playwright script for every site. The skill emphasizes trusting match metadata instead of guessing, handling stale element refs, and using network capture when the task needs evidence of what fired. It is not for publishing reusable site adapters; that workflow belongs to opencli-adapter-author. Allowed tooling is scoped to Bash(opencli:* ) plus normal file edits when outputs need persisting.11.6kinstalls3Opencli AutofixOpenCLI AutoFix is an agent skill that turns a failed opencli command into a disciplined repair workflow instead of a one-line error dump. When a third-party site shifts markup or APIs, solo builders depend on thin CLI adapters; this skill tells the agent to capture a trace, read adapterSourcePath from the trace summary, patch that single adapter file—whether under clis/ in a repo or ~/.opencli/clis/ for npm installs—and retry with a three-round budget. Safety gates matter: authentication and browser connectivity failures stop immediately, and CAPTCHA or rate limits are treated as environmental, not code bugs. Allowed tooling spans opencli and gh bash, plus read/write/edit on the adapter. After a verified fix, it can file an upstream issue so the community benefits. It is for operators and integrators who already run OpenCLI, not for greenfield CLI framework work inside the core package.11kinstalls4Smart SearchSmart-search teaches coding agents how to route opencli when the user never said which website to hit. Instead of firing grok, doubao, and gemini in parallel, the agent runs `opencli list -f yaml`, confirms which of the three AI sources exist in the registry, inspects help for the site and subcommand, then commits to one default lane matched to the question. Grok fits live English social and X-centric discourse; doubao fits Chinese domestic hotspots and lifestyle or Byte-ecosystem angles; gemini fits broad English web synthesis with pointers toward google, wikipedia, arxiv, or stackoverflow. The readme embeds worked query patterns—time windows, platform hints, and desired output form—so solo builders get faster, cheaper research passes during idea validation, competitive scans, growth messaging checks, or operational fact-finding without drowning in duplicate AI summaries.10.4kinstalls5Opencli Adapter AuthorOpenCLI Adapter Author teaches solo builders how to add commands to the OpenCLI ecosystem using the canonical single cli({...}) call pattern. The skill is a template walkthrough anchored on eastmoney convertible.js: site and name declaration, PUBLIC strategy, domain binding, argument schema with defaults and help strings, column maps, and fetch/transform logic. It explicitly warns that limit clamping and CliError codes in the sample are grandfathered—new work must follow typed-errors.md and §3 validation rules from 2026-05 lint baselines. You install it when you need consistent, discoverable CLIs over HTTP data sources without reinventing registry conventions. Intermediate complexity fits developers comfortable with ES modules and REST-shaped market data. Deliverable is a production-shaped adapter file ready for registry loading, not a deployed service by itself.8.2kinstalls6Opencli Exploreropencli-explorer is a procedural skill for solo builders who want a maintainable CLI against a website that does not ship a public API doc. It walks you from opening the target URL in OpenCLI’s browser tooling through observing network traffic after real UI interactions, choosing an authentication cascade, implementing adapters with the TypeScript cli API, and running verification. The guide explicitly branches: one-shot page commands go to opencli-oneshot, automated `opencli generate` is the fast path when it works, and this explorer doc is the fallback when discovery is messy or you need multiple commands. It stresses that agents must use the browser because lazy-loaded data only appears after user-like navigation. Whether you publish a PR to the main clis tree or keep a private adapter under the home directory, the same discovery discipline applies. Complexity is advanced because auth, anti-bot behavior, and endpoint stability vary by site.4kinstalls7Opencli Oneshotopencli-oneshot is a procedural skill for solo builders who need one CLI command fast from a live web app. You supply a URL and a short goal—such as listing Twitter Lists—and the agent follows four steps: open the page with OpenCLI browser capture, wait and inspect filtered JSON network traffic, lock one endpoint with --detail, then author and test a TypeScript adapter. It deliberately stays narrow; the readme lists hard stops where you must switch to opencli-explorer, including anti-bot signatures, Pinia store actions, multi-command sites, or empty network traces that need bundle analysis. Category placement is CLI and terminal work with heavy integration subtext. Best for agents already using the OpenCLI toolchain who want repeatable adapter generation without a full site crawl. Not for enterprise-wide CLI suites or flows that need deep exploration on the first pass.3.9kinstalls8Opencli OperateOpenCLI Operate is an agent skill for controlling Chrome through a CLI while reusing whatever sessions you already have open—ideal when a solo builder wants the agent to operate SaaS dashboards, internal tools, or staging sites the way a human would. It emphasizes structured page introspection with `state` (fast, token-cheap DOM indices) instead of vision screenshots, and native `click`/`type`/`select` instead of brittle `eval` clicks. You install it when web automation must respect real auth cookies, when you are validating flows during build, when you need repeatable ship-time checks, or when you are doing light production ops that still live in the browser. The skill encodes a small but strict discipline: doctor first, state after every navigation change, verify typed fields, and chain commands aggressively. It is integration-heavy tooling—not a replacement for Playwright test suites at scale—but it shines for agent-in-the-loop tasks where setup time and API keys are the bottleneck.1.9kinstalls9Opencli Repairopencli-repair is an agent skill for solo builders who rely on OpenCLI adapters to drive websites from the terminal. When a command fails because a site changed selectors, response shapes, or login flows, the skill prescribes a diagnostic-first workflow: verify connectivity with opencli doctor, re-run the failing command with OPENCLI_DIAGNOSTIC=1 to capture structured RepairContext, then use browser automation to see what changed and edit the adapter in place. It targets Build/integrations while you are extending CLI coverage, and Operate/iterate when daily scripts suddenly return EMPTY_RESULT or PAGE_CHANGED. The skill is explicitly error-code driven so the agent does not guess—each code hints whether DOM, network, or adapter logic regressed. Outcome is a patched adapter and a passing opencli invocation, not a one-off manual scrape.664installs10Opencli Browser Sitemapopencli-browser-sitemap is an agent skill for teams that automate real websites with opencli browser when a sitemap bundle is available or requested. It tells the agent to pull the smallest relevant sitemap files lazily—site orientation, the current page doc, and the workflow for the user goal—then follow the workflow’s best path, including dedicated opencli adapters when listed. After every state-changing action it re-runs browser state and checks the workflow state signature so navigation stays grounded in what is on screen, not stale URLs. When adapters fail or are missing it falls back to explicit browser steps, and it marks sitemap entries stale when reality disagrees. Solo builders shipping scrapers, internal ops bots, or agent-driven QA against complex SPAs install it to cut random exploration while keeping safety boundaries around Bash(opencli:*), Read, Edit, Write, and Grep.127installs