AI Web Check performs an evidence-first, non-mutating assessment of website usability and agent readiness.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ai-web-check
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install ai-web-check using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is AI Web Check?
AI Web Check is an Openclaw Skills utility for evaluating whether a website is usable by AI agents without relying on page copy alone. Its bundled CLI scans a target site and records check-specific evidence in Markdown reports.
The skill also supports bounded, read-only browser journeys based on same-origin GET pages. It is intentionally non-mutating: it does not fill forms, authenticate, submit data, send messages, make payments, delete content, or follow cross-origin journeys. Openclaw Skills users should treat NOT_TESTED results as unknown rather than passing, and understand that a PASS applies only to the named check and its recorded evidence.
AI Web Check Use Cases
- Assess whether a website is ready for AI-agent navigation.
- Run a repeatable usability scan before integrating a site with an AI workflow.
- Inspect a bounded set of same-origin pages through a read-only browser journey.
- Generate evidence-backed Markdown reports for engineering or product review.
- Identify candidates for structured machine actions, human review flows, state consistency checks, idempotency testing, or version-conflict analysis.
- Distinguish verified behavior from unsupported assumptions when evaluating websites with Openclaw Skills.
How AI Web Check Works
- Choose a public website URL to evaluate.
- Run the bundled CLI scan command to inspect the site and generate evidence for supported checks.
- Optionally define a JSON journey containing same-origin GET pages for a bounded, read-only navigation flow.
- Run the journey command against the target URL.
- Review the Markdown output and the evidence associated with each named check.
- Interpret PASS only as evidence for that specific check; treat NOT_TESTED as unknown, not as a pass.
- Identify runtime properties that require additional validation, including structured machine actions, human review, cross-entry state consistency, idempotency, version conflicts, and truthful state reporting.
AI Web Check Setup
- Open the AI Web Check skill directory containing the bundled
src/cli.mjsentry point. - Ensure Node.js is available in the execution environment.
- Run a basic website scan:
node src/cli.mjs scan https://example.com --format markdown --out report.md
- For a bounded read-only journey, create a JSON file such as
journey.jsoncontaining same-origin GET pages. - Run the journey assessment:
node src/cli.mjs journey https://example.com --journey journey.json --format markdown
- Review the generated Markdown report and preserve its evidence when sharing results. No authentication, form submission, payment, deletion, messaging, or cross-origin journey setup is supported by the v0.1 runner.
AI Web Check Data Schema & Taxonomy
Inputs
- Target URL: The website supplied to
scanorjourney. - Journey definition: A JSON file listing same-origin pages to request through a bounded GET-only journey.
- Output format: Markdown, selected with
--format markdown. - Output path: Optional report destination supplied with
--out report.mdfor scans.
Assessment results
| Result | Meaning |
|---|---|
PASS |
The named check has recorded supporting evidence; it is not vendor certification. |
NOT_TESTED |
The check is unknown and must not be interpreted as a pass. |
Generated artifacts
- Scan reports can be written to a Markdown file such as
report.md. - Journey output is emitted in Markdown when
--format markdownis specified. - Evidence is organized around named checks rather than a blanket certification.
The v0.1 data model does not itself certify structured machine actions, human review flows, cross-entry state consistency, idempotency or version conflicts, or state truthfulness. These properties require additional evidence and runtime validation.
AI Web Check Advanced Features
- Evidence-first reporting instead of inferring success from page copy.
- Non-mutating execution designed for safe website reconnaissance.
- Bounded browser journeys using same-origin GET requests.
- Markdown output suitable for engineering reviews and audit trails.
- Explicit PASS and NOT_TESTED semantics that prevent unknown behavior from being treated as verified.
- Safety boundaries that block authentication, form filling, submissions, payments, deletions, outbound sending, and cross-origin journeys.
- Candidate discovery for deeper AI-readiness validation without claiming runtime certification.
- A focused workflow for Openclaw Skills integrations that need conservative, reproducible website checks.
SKILL.md
Loading
Related Openclaw Skills
Hermes Browser on Fedora Atomic
A practical troubleshooting and configuration skill that makes Hermes browser automation reliable on Fedora Atomic and bootc systems.
Device Connectivity
Device Connectivity pairs Android or iOS phones with Linux desktops through the KDE Connect protocol for seamless cross-device control and data sharing.
Hermes Web Tools
Hermes Web Tools provides a read-only workflow for configuring, selecting, and troubleshooting Hermes web_search and web_extract backends.
Starship Prompt
A complete developer reference for installing, configuring, customizing, and troubleshooting the fast, cross-shell Starship prompt.
Zhongkui Guard
Zhongkui Guard scans AI-related content for suspicious commands, redacts sensitive data, and evaluates configured model APIs with evidence-based reports.
Client Follow-up Plan
Convert client meeting notes, chat excerpts, and visit records into a structured, evidence-based follow-up plan with owners, deadlines, risks, and completion criteria.
Featured*
