Stop Blind Clicking: Sitemap-Guided Browser Agents

Where this fits

You're driving a site with opencli browser, and the analyzer reports sitemap.available: true. Without a plan, the agent guesses: opens links, hits dead ends, and makes five clicks to reach a page the sitemap already described. This skill gives agents a lazy sitemap consumption loop: inspect the current state, read only the relevant SITE.md, pages/<page-id>.md, and workflows/<task-id>.md files, then follow the workflow's best path. It's for developers who want browser agents to navigate with context instead of by trial and error.

Why agents benefit

• Reads the smallest relevant sitemap files on demand, so startup stays cheap and context stays small.
• Chooses adapter or browser fallback paths when the sitemap is incomplete or the target page is missing.
• Resumes from state signatures after a browser session changes, avoiding repeated setup.
• Marks stale sitemap entries without trusting them over live browser state.
• Prefers the workflow's best path, reducing blind clicking and wasted tool calls.

Practical scenarios

Onboarding to an unfamiliar docs site

An agent is asked to find the API endpoint for creating a webhook. The sitemap includes workflows/webhook-create.md; the skill reads that file and follows the named path, instead of crawling the docs until it gets lucky.

Resuming after a browser session interrupts

The agent paused mid-task, and the page changed. The skill checks opencli browser <session> state, compares the state signature, and reuses the relevant sitemap page rather than restarting the navigation from scratch.

Sitemap drift versus live browser state

The sitemap says a form lives at /old-path, but the live page has already moved to /new-path. The agent trusts the rendered DOM, records the stale entry, and continues without failing or forcing the old URL.

Add it to your agent workflow

Install the skill with:

npx skills add jackwener/OpenCLI --skill opencli-browser-sitemap

Then, when a session reports sitemap.available: true, let the skill drive the next steps:

opencli browser open --url https://example.com
opencli browser analyze --session s1
# sitemap.available: true -> opencli-browser-sitemap
opencli browser s1 state

The skill decides which sitemap files matter and how to reconcile them with live browser state.

Compare before adopting

Before adding this skill, compare it against your current browser-agent loop. Look at maintenance freshness: is the OpenCLI skill kept up to date with the CLI's adapter and state signatures? Check alternatives: a generic scraping skill may be broader, but this one is specialized for OpenCLI sitemap-driven navigation. Finally, measure workflow fit: if your agents never need sitemap context, the skill adds little; if they routinely navigate unknown sites, it removes a major failure mode.

Why it is worth tracking

The skill is tied to OpenCLI's momentum; the project's 28.8k stars suggest an active ecosystem, and this skill addresses a real gap by converting sitemaps from static data into an agent navigation policy. It is worth tracking because it models an increasingly common pattern: treat prior knowledge as advisory, validate against live state, and mark stale signals instead of trusting them. For a closer look at the skill's design decisions, see the OpenAgentSkill listing at https://www.openagentskill.com/skills/jackwener-opencli-browser-sitemap. Evaluate it again when your agents start spending more time on multi-step browser workflows.