OpenCLI Browser Sitemap Navigator
Drive websites efficiently using sitemap context to avoid blind clicking and adhere to live browser reality.
The skill explicitly instructs agents to trust live browser state over sitemap, limiting writes to local overrides, showing data-flow transparency and least privilege. But it allows Bash execution of opencli commands without explicit user confirmation or command disclosure, and relies on unverified external tools. Deductions: permissions, confirmation, and recovery mechanisms are incomplete, and data-flow disclosure is not fully declared.
The skill logic is self-consistent with clear step-by-step procedures and failure handling (e.g., Trust Reality rule). However, it lacks self-tests or explicit error message examples, and static evaluation cannot verify key paths, so the 10-point cap is not reached. Deductions: lack of executable test evidence and error feedback details, static evaluation limitations.
The skill has clear usage scenarios and trigger conditions (sitemap.available: true), and declares capability boundaries (does not overwrite global seed files). But it does not explicitly state non-fitting environments, and depends on external tool availability without assessing mainland-China network reachability. Deductions: boundary and environment fit information is incomplete.
The skill provides clear steps and usage guidance, but lacks version info, changelog, known limitations, and troubleshooting sections. License is in repository but not explicitly referenced in skill. Deductions: information architecture and governance are incomplete.
The skill describes expected outputs (path chosen, checkpoint, sitemap health) but provides no practical examples or verified results, and marginal value is limited. Static evaluation cannot confirm task completion. Deductions: lack of output evidence, static evaluation limitations.
Repository has CI workflows and test files, but they do not cover the skill's key path (sitemap consumption). No independent execution evidence, only source files. Deductions: evidence types are singular and coverage is thin, static evaluation limitations.
- This skill depends on the OpenCLI toolchain, whose installation and operation may involve untrusted external services; verify security yourself.
- The skill emphasizes local file priority but does not clarify how to handle sensitive data (e.g., login cookies); avoid writing sensitive info into sitemaps or logs.
- In mainland-China networks, reachability of Chrome Web Store and GitHub may be limited, affecting skill initialization and usage.
- No independent testing verification; test the skill behavior in a controlled environment before real use.
What does this skill do, and when should you use it?
This skill equips AI agents to use site sitemaps as prior knowledge while navigating with OpenCLI's browser commands. It guides a lazy consumption loop: read only the smallest relevant sitemap files, prefer adapter best paths with browser fallbacks, and refresh state after each action. The skill emphasizes trust in reality over the sitemap, marking stale entries in a local overlay and writing back adapter health issues. It integrates with OpenCLI sessions and requires the installed OpenCLI CLI and a connected browser.
Reads sitemap files (local overlay first, then global seed) and current browser state via opencli browser <session> state. Chooses adapter best paths or browser fallback workflows, compares state signatures after actions, and probes reality when mismatches occur. Writes local stale notes and adapter health updates to the local overlay, never editing global seeds during tasks. Reports chosen path, checkpoint, and sitemap health in output.
- An agent navigating a complex website with a sitemap available to reduce blind clicking.
- A user asks an agent to use the sitemap for a site with known workflows, so the agent follows the best path and falls back to browser actions if needed.
- A developer debugging a broken adapter uses the skill to mark it suspect and switch to browser fallback.
- A content writer extracting data from a site uses sitemap context to locate relevant pages quickly without loading the whole sitemap.
What are this skill's strengths and limitations?
- Reduces blind clicking by using sitemap prior knowledge.
- Trusts live browser state, marking stale sitemap entries.
- Lazy loading of sitemap files avoids context bloat.
- Provides adapter health write-back for continuous improvement.
- Clear output discipline for reporting.
- Requires OpenCLI and its browser bridge to be installed and connected.
- Only for AI agents that can execute shell commands and have OpenCLI available.
- Local overlay notes are agent-specific and not shared across agents unless manually synced.
- No built-in test suite; relies on real browser sessions.
How do you install this skill?
Install OpenCLI: use OpenCLIApp (macOS/Windows) or npm install -g @jackwener/opencli (requires Node.js >= 20). Install the Browser Bridge extension from the Chrome Web Store or manually. Then add the skill to your AI agent with npx skills add jackwener/opencli --skill opencli-browser-sitemap. Ensure a sitemap directory exists for the site (local overlay under ~/.opencli/sites/<site>/sitemap/ or global seed under sitemaps/<site>/).
How do you use this skill?
Trigger the skill when opencli browser open or analyze reports sitemap.available: true, or ask your agent to 'Use the sitemap to navigate this website'. The agent will run opencli browser <session> state, read relevant sitemap files, follow the workflow's best path or fallback, and refresh state after actions. Use natural language like 'Help me find this product on the site using the sitemap.'
How does this skill compare with similar options?
N/A