ClawHub skill

Honest eyes for your OpenClaw agent

An agent that can shell out to curl can already fetch a URL. What it cannot do is tell you whether the fetch actually worked. This skill adds six operations over the EnConvert API, and every page read comes back with a render_quality score, so a bot wall or an empty app shell arrives flagged instead of being summarised as though it were the page.

Install the skill
openclaw skills install @enconvert/enconvert
Read the documentationView the source
The gap

Why curl alone is not enough

OpenClaw agents are good at shell commands. The web is bad at being read by them. Four things break in practice, and the skill covers each.

What breaksWhat the skill does

A page that renders in JavaScript

Perceive URL runs a real browser and returns the settled page as markdown, not the empty shell that curl sees.

A Word, PowerPoint or EPUB file

Convert File to Markdown reads formats an agent has no parser for, and returns text it can reason about.

Knowing whether a read succeeded

Every page read carries render_quality from 0.0 to 1.0. A challenge page scores low, so the agent can retry or stop instead of hallucinating from a cookie banner.

Reading a whole site, not one page

Discover URLs lists a site by sitemap or crawl, so the agent picks what to read instead of guessing at URLs.

Operations

Six operations, one instruction file

A ClawHub skill is a single markdown file the agent reads and acts on. There is no runtime to install, no SDK to pin, and no tool schema to register. Each operation is a documented curl call against the same REST API every other EnConvert client uses.

  1. Perceive URL

    Render one page into markdown, HTML, a screenshot, a PDF, links, images or structured data. Artifacts come back as short-lived signed URLs; structured data comes back inline.

    POST /v2/perceive

  2. Web Search

    Search the web, news, images, scholar, patents or maps and get ranked results with titles, URLs and snippets to feed the next step.

    POST /v2/lookup

  3. Discover URLs

    Enumerate a site by sitemap, crawl or both, with include and exclude patterns, before spending a single page read.

    POST /v2/discover

  4. Extract Structured

    Give a JSON schema or a flat field map and get typed data back per URL, from an explicit list or from a site the API crawls for you.

    POST /v2/distill

  5. Convert File to Markdown

    Turn a document, spreadsheet, presentation, PDF or EPUB into clean markdown, ready to summarise or embed.

    POST /v1/convert/anything-to-markdown

  6. Convert File to PDF

    Convert a document, image, HTML or text file to PDF when the agent needs to hand a real file back to a person.

    POST /v1/convert/anything-to-pdf
Setup

One command and one secret

The skill ships no key and calls nothing but the EnConvert API. You supply your own private key, and the agent reads it from the environment.

Install from ClawHub

Run the install command in your agent workspace. The skill is public, MIT-0, and published under the @enconvert publisher.

Create a private key

Generate a key in the EnConvert dashboard. Private keys start with sk_. Public pk_ keys are for browser widgets and are rejected here.

Expose it as ENCONVERT_API_KEY

Put the key in the environment the agent runs in. The skill validates it against a whoami call before doing any real work, so a bad key fails immediately rather than midway through a task.

Ask the agent for something

No wiring, no config file. Ask it to read a page, research a topic or convert a file, and it picks the operation from the instructions.

If ENCONVERT_API_KEY or curl is missing, OpenClaw filters the skill out at load time and the agent behaves as though it was never installed. There is no error message to read, so check both first if the agent seems unaware of the skill.

In practice

What agents actually do with it

Four tasks that need no prompt engineering. The agent reads the skill and chains the operations itself.

"Read this page and summarise it"

  • Perceive URL

Perceive returns settled markdown with a quality score, so the summary is of the page rather than of a consent dialog.

"Research this topic and cite sources"

  • Web Search
  • Perceive URL

Search finds candidates, then perceive reads the promising ones. The agent cites URLs it actually read.

"Pull pricing from every competitor page"

  • Discover URLs
  • Extract Structured

Discover lists the site, then extract returns typed fields per URL against one schema.

"Turn this contract into notes"

  • Convert File to Markdown

Convert reads the Word or PDF file the agent has no parser for, and hands back markdown.

Frequently asked questions

ClawHub is the skill and plugin registry for OpenClaw agents. A skill is a markdown instruction file an agent reads and executes, so installing one teaches the agent a capability rather than adding a dependency.
Yes. The skill is free and carries no key of its own. You supply a private key from your own dashboard, and usage counts against your own plan, exactly as it does through the CLI, the SDKs or the API.
Almost always a missing ENCONVERT_API_KEY or a missing curl. OpenClaw filters skills at load time when their requirements are unmet, and it does so silently, so the agent genuinely does not know the skill exists.
No. Two fetches deliberately carry no key: downloading a source file before a conversion, and fetching a signed output URL. Both are third-party or pre-authenticated, so sending the key would leak it for no benefit.
A score from 0.0 to 1.0 describing how well a page actually rendered. A blocked, empty or bot-walled page scores low. The skill instructs the agent to treat a low score as a failed read rather than as content, which is the difference between a wrong answer and a retry.
The MCP server exposes the same operations as real registered tools to any client that speaks the Model Context Protocol. The ClawHub skill is for OpenClaw agents specifically and works through curl. Use whichever your agent supports.

Give your agent honest eyes

Install the skill, set one environment variable, and the next thing you ask your agent to read comes back with a score attached.

Read the documentationClawHub