zeroclaw-labs/zeroclaw · error

unsupported URL scheme: {}

Error message

unsupported URL scheme: {}

What it means

browser_delegate validates every URL it touches — both explicit task URLs and URLs extracted from free-form task text by a regex — through reqwest::Url parsing followed by a scheme allowlist: only http and https pass. Any other scheme (ftp, file, ws, chrome, javascript, ...) is rejected before the browser CLI subprocess sees it.

Source

Thrown at crates/zeroclaw-tools/src/browser_delegate.rs:90

    fn validate_url(&self, url: &str) -> anyhow::Result<()> {
        let parsed = url.parse::<reqwest::Url>().map_err(|e| {
            ::zeroclaw_log::record!(
                WARN,
                ::zeroclaw_log::Event::new(module_path!(), ::zeroclaw_log::Action::Reject)
                    .with_outcome(::zeroclaw_log::EventOutcome::Failure)
                    .with_attrs(::serde_json::json!({
                        "url": url,
                        "error": format!("{}", e),
                    })),
                "browser_delegate: invalid URL"
            );
            anyhow::Error::msg(format!("invalid URL '{}': {}", url, e))
        })?;

        // Only allow http/https schemes
        let scheme = parsed.scheme();
        if scheme != "http" && scheme != "https" {
            anyhow::bail!("unsupported URL scheme: {}", scheme);
        }

        let domain = parsed.host_str().unwrap_or("").to_string();

        if domain.is_empty() {
            anyhow::bail!("URL has no host: {}", url);
        }

        // Check blocked domains first (deny takes precedence)
        for blocked in &self.config.blocked_domains {
            if domain_matches(&domain, blocked) {
                anyhow::bail!("domain '{}' is blocked by browser_delegate policy", domain);
            }
        }

        // If allowed_domains is non-empty, it acts as an allowlist
        if !self.config.allowed_domains.is_empty() {
            let allowed = self

View on GitHub (pinned to 88bb9c8533)

Solutions

1. Remove or rewrite non-http(s) links in the task text (use https:// mirrors)
2. If another scheme is genuinely needed, handle it outside browser_delegate — the allowlist is a security boundary
3. Pre-scan task text with the same https?:// extraction and reject before submitting

Example fix

// before
task = "Download ftp://mirror.example.org/pkg.tar.gz"
// after
task = "Download https://mirror.example.org/pkg.tar.gz"

for m in regex::Regex::new(r"https?://[^\s\)\]\},\"'`<>")
    .unwrap()
    .find_iter(&task)
{
    let parsed: reqwest::Url = m.as_str().parse()?;
    if parsed.scheme() != "http" && parsed.scheme() != "https" {
        return Err(format!("non-http(s) URL in task: {}", m.as_str()));
    }
}

match delegate.execute(args).await {
    Ok(res) if res.success => { /* ... */ }
    Ok(res) => {
        if res.error.as_deref().unwrap_or_default().contains("unsupported URL scheme") {
            // strip or rewrite the offending link from the task text
        }
    }
    Err(e) => return Err(e),
}

When it happens

Trigger: A task containing `ftp://mirror.example.com/file` or `file:///etc/passwd`. Note that validate_task_urls scans the entire prompt with a https?:// regex, so a scheme-violating link buried in prose also triggers this check.

Common situations: Agents pasting README links that mix ftp/file schemes into the delegated task; users expecting desktop-browser scheme tolerance; test fixtures using about:blank or chrome:// URLs.

Related errors

• Custom model_provider `{prefix}:<url>` requires a URL beginn
• OpenAI Codex endpoint override must use http:// or https://
• URL has no host: {}
• Only http:// or https:// URLs are allowed
• Only http:// and https:// URLs are allowed

AI-assisted analysis of zeroclaw-labs/zeroclaw@88bb9c8533 (2026-08-23). Data as JSON: /api/errors/3c010f8061bc1a79. Report an issue: GitHub.