zeph_tools/

scrape.rs

// SPDX-FileCopyrightText: 2026 Andrei G <bug-ops>
// SPDX-License-Identifier: MIT OR Apache-2.0

//! Web scraping executor with SSRF protection and domain policy enforcement.
//!
//! Exposes two tools to the LLM:
//!
//! - **`web_scrape`** — fetches a URL and extracts elements matching a CSS selector.
//! - **`fetch`** — fetches a URL and returns the raw response body as UTF-8 text.
//!
//! Both tools enforce:
//!
//! - HTTPS-only URLs (HTTP and other schemes are rejected).
//! - DNS resolution followed by a private-IP check to prevent SSRF.
//! - Optional domain allowlist and denylist from [`ScrapeConfig`].
//! - Configurable timeout and maximum response body size.
//! - Redirect following is disabled to prevent open-redirect SSRF bypasses.

use std::net::SocketAddr;
use std::sync::Arc;
use std::sync::atomic::{AtomicU64, Ordering};
use std::time::{Duration, Instant};

use schemars::JsonSchema;
use serde::Deserialize;
use url::Url;

use zeph_common::ToolName;

use zeph_sanitizer::IpiFilter;

use crate::audit::{AuditEntry, AuditLogger, AuditResult, EgressEvent, chrono_now};
use crate::config::{EgressConfig, ScrapeConfig};
use crate::executor::{
    ClaimSource, ToolCall, ToolError, ToolExecutor, ToolOutput, deserialize_params,
};
use crate::net::{check_domain_policy, validate_url};

/// Strips userinfo (`user:pass@`) and sensitive query params from a URL for safe logging.
///
/// Returns the sanitized URL string; falls back to the original if parsing fails.
fn redact_url_for_log(url: &str) -> String {
    let Ok(mut parsed) = Url::parse(url) else {
        return url.to_owned();
    };
    // Remove userinfo.
    let _ = parsed.set_username("");
    let _ = parsed.set_password(None);
    // Strip query params whose names suggest secrets (token, key, secret, password, auth, sig).
    let sensitive = [
        "token", "key", "secret", "password", "auth", "sig", "api_key", "apikey",
    ];
    let filtered: Vec<(String, String)> = parsed
        .query_pairs()
        .filter(|(k, _)| {
            let lower = k.to_lowercase();
            !sensitive.iter().any(|s| lower.contains(s))
        })
        .map(|(k, v)| (k.into_owned(), v.into_owned()))
        .collect();
    if filtered.is_empty() {
        parsed.set_query(None);
    } else {
        let q: String = filtered
            .iter()
            .map(|(k, v)| format!("{k}={v}"))
            .collect::<Vec<_>>()
            .join("&");
        parsed.set_query(Some(&q));
    }
    parsed.to_string()
}

#[derive(Debug, Deserialize, JsonSchema)]
struct FetchParams {
    /// HTTPS URL to fetch
    url: String,
}

#[derive(Debug, Deserialize, JsonSchema)]
struct ScrapeInstruction {
    /// HTTPS URL to scrape
    url: String,
    /// CSS selector
    select: String,
    /// Extract mode: text, html, or attr:<name>
    #[serde(default = "default_extract")]
    extract: String,
    /// Max results to return
    limit: Option<usize>,
}

fn default_extract() -> String {
    "text".into()
}

#[derive(Debug)]
enum ExtractMode {
    Text,
    Html,
    Attr(String),
}

impl ExtractMode {
    fn parse(s: &str) -> Self {
        match s {
            "text" => Self::Text,
            "html" => Self::Html,
            attr if attr.starts_with("attr:") => {
                Self::Attr(attr.strip_prefix("attr:").unwrap_or(attr).to_owned())
            }
            _ => Self::Text,
        }
    }
}

/// Extracts data from web pages via CSS selectors.
///
/// Handles two invocation paths:
///
/// 1. **Legacy fenced blocks** — detects ` ```scrape ` blocks in the LLM response, each
///    containing a JSON scrape instruction object. Dispatched via [`ToolExecutor::execute`].
/// 2. **Structured tool calls** — dispatched via [`ToolExecutor::execute_tool_call`] for
///    tool IDs `"web_scrape"` and `"fetch"`.
///
/// # Security
///
/// - Only HTTPS URLs are accepted. HTTP and other schemes return [`ToolError::InvalidParams`].
/// - DNS is resolved synchronously and each resolved address is checked against
///   [`crate::net::is_private_ip`]. Private addresses are rejected to prevent SSRF.
/// - HTTP redirects are disabled (`Policy::none()`) to prevent open-redirect bypasses.
/// - Domain allowlists and denylists from config are enforced before DNS resolution.
///
/// # Example
///
/// ```rust,no_run
/// use zeph_tools::{WebScrapeExecutor, ToolExecutor, ToolCall, ScrapeConfig};
/// use zeph_common::ToolName;
///
/// # async fn example() {
/// let executor = WebScrapeExecutor::new(&ScrapeConfig::default());
///
/// let call = ToolCall {
///     tool_id: ToolName::new("fetch"),
///     params: {
///         let mut m = serde_json::Map::new();
///         m.insert("url".to_owned(), serde_json::json!("https://example.com"));
///         m
///     },
///     caller_id: None,
///     context: None,
///     tool_call_id: String::new(),
///     skill_name: None,
/// };
/// let _ = executor.execute_tool_call(&call).await;
/// # }
/// ```
#[derive(Debug)]
pub struct WebScrapeExecutor {
    timeout: Duration,
    max_body_bytes: usize,
    allowed_domains: Vec<String>,
    denied_domains: Vec<String>,
    audit_logger: Option<Arc<AuditLogger>>,
    egress_config: EgressConfig,
    egress_tx: Option<tokio::sync::mpsc::Sender<EgressEvent>>,
    egress_dropped: Arc<AtomicU64>,
    /// IPI filter applied to every fetched response body before returning to callers.
    ipi_filter: IpiFilter,
}

impl WebScrapeExecutor {
    /// Create a new `WebScrapeExecutor` from configuration.
    ///
    /// No network connections are made at construction time.
    #[must_use]
    pub fn new(config: &ScrapeConfig) -> Self {
        Self {
            timeout: Duration::from_secs(config.timeout),
            max_body_bytes: config.max_body_bytes,
            allowed_domains: config.allowed_domains.clone(),
            denied_domains: config.denied_domains.clone(),
            audit_logger: None,
            egress_config: EgressConfig::default(),
            egress_tx: None,
            egress_dropped: Arc::new(AtomicU64::new(0)),
            ipi_filter: IpiFilter::new(config.ipi_filter_threshold),
        }
    }

    /// Attach an audit logger. Each tool invocation will emit an [`AuditEntry`].
    #[must_use]
    pub fn with_audit(mut self, logger: Arc<AuditLogger>) -> Self {
        self.audit_logger = Some(logger);
        self
    }

    /// Configure egress event logging.
    #[must_use]
    pub fn with_egress_config(mut self, config: EgressConfig) -> Self {
        self.egress_config = config;
        self
    }

    /// Attach the egress telemetry channel sender and drop counter.
    ///
    /// Events are sent via [`tokio::sync::mpsc::Sender::try_send`] — the executor
    /// never blocks waiting for capacity.
    #[must_use]
    pub fn with_egress_tx(
        mut self,
        tx: tokio::sync::mpsc::Sender<EgressEvent>,
        dropped: Arc<AtomicU64>,
    ) -> Self {
        self.egress_tx = Some(tx);
        self.egress_dropped = dropped;
        self
    }

    /// Returns a clone of the egress drop counter, for use in the drain task.
    #[must_use]
    pub fn egress_dropped(&self) -> Arc<AtomicU64> {
        Arc::clone(&self.egress_dropped)
    }

    fn build_client(&self, host: &str, addrs: &[SocketAddr]) -> reqwest::Client {
        let mut builder = reqwest::Client::builder()
            .timeout(self.timeout)
            .redirect(reqwest::redirect::Policy::none());
        builder = builder.resolve_to_addrs(host, addrs);
        builder.build().unwrap_or_default()
    }
}

impl ToolExecutor for WebScrapeExecutor {
    fn tool_definitions(&self) -> Vec<crate::registry::ToolDef> {
        use crate::registry::{InvocationHint, ToolDef};
        vec![
            ToolDef {
                id: "web_scrape".into(),
                description: "Extract structured data from a web page using CSS selectors.\n\nONLY call this tool when the user has explicitly provided a URL in their message, or when a prior tool call returned a URL to retrieve. NEVER construct, guess, or infer a URL from entity names, brand knowledge, or domain patterns.\n\nParameters: url (string, required) - HTTPS URL; select (string, required) - CSS selector; extract (string, optional) - \"text\", \"html\", or \"attr:<name>\"; limit (integer, optional) - max results\nReturns: extracted text/HTML/attribute values, one per line\nErrors: InvalidParams if URL is not HTTPS or selector is empty; Timeout after configured seconds; connection/DNS failures".into(),
                schema: schemars::schema_for!(ScrapeInstruction),
                invocation: InvocationHint::FencedBlock("scrape"),
                output_schema: None,
                server_id: None,
            },
            ToolDef {
                id: "fetch".into(),
                description: "Fetch a URL and return the response body as plain text.\n\nONLY call this tool when the user has explicitly provided a URL in their message, or when a prior tool call returned a URL to retrieve. NEVER construct, guess, or infer a URL from entity names, brand knowledge, or domain patterns. If no URL is present in the conversation, do not call this tool.\n\nParameters: url (string, required) - HTTPS URL to fetch\nReturns: response body as UTF-8 text, truncated if exceeding max body size\nErrors: InvalidParams if URL is not HTTPS; Timeout; SSRF-blocked private IPs; connection failures".into(),
                schema: schemars::schema_for!(FetchParams),
                invocation: InvocationHint::ToolCall,
                output_schema: None,
                server_id: None,
            },
        ]
    }

    async fn execute(&self, response: &str) -> Result<Option<ToolOutput>, ToolError> {
        let blocks = extract_scrape_blocks(response);
        if blocks.is_empty() {
            return Ok(None);
        }

        let mut outputs = Vec::with_capacity(blocks.len());
        #[allow(clippy::cast_possible_truncation)]
        let blocks_executed = blocks.len() as u32;

        for block in &blocks {
            let instruction: ScrapeInstruction = serde_json::from_str(block).map_err(|e| {
                ToolError::Execution(std::io::Error::new(
                    std::io::ErrorKind::InvalidData,
                    e.to_string(),
                ))
            })?;
            let correlation_id = EgressEvent::new_correlation_id();
            let start = Instant::now();
            let scrape_result = self
                .scrape_instruction(&instruction, &correlation_id, None, None)
                .await;
            #[allow(clippy::cast_possible_truncation)]
            let duration_ms = start.elapsed().as_millis() as u64;
            match scrape_result {
                Ok(output) => {
                    self.log_audit(
                        "web_scrape",
                        &redact_url_for_log(&instruction.url),
                        AuditResult::Success,
                        duration_ms,
                        None,
                        None,
                        None,
                        Some(correlation_id),
                    )
                    .await;
                    outputs.push(output);
                }
                Err(e) => {
                    let audit_result = tool_error_to_audit_result(&e);
                    self.log_audit(
                        "web_scrape",
                        &redact_url_for_log(&instruction.url),
                        audit_result,
                        duration_ms,
                        Some(&e),
                        None,
                        None,
                        Some(correlation_id),
                    )
                    .await;
                    return Err(e);
                }
            }
        }

        Ok(Some(ToolOutput {
            tool_name: ToolName::new("web-scrape"),
            summary: outputs.join("\n\n"),
            blocks_executed,
            filter_stats: None,
            diff: None,
            streamed: false,
            terminal_id: None,
            locations: None,
            raw_response: None,
            claim_source: Some(ClaimSource::WebScrape),
            ..Default::default()
        }))
    }

    #[cfg_attr(
        feature = "profiling",
        tracing::instrument(name = "tools.scrape.fetch", skip_all)
    )]
    async fn execute_tool_call(&self, call: &ToolCall) -> Result<Option<ToolOutput>, ToolError> {
        match call.tool_id.as_str() {
            "web_scrape" => {
                let instruction: ScrapeInstruction = deserialize_params(&call.params)?;
                let correlation_id = EgressEvent::new_correlation_id();
                let start = Instant::now();
                let result = self
                    .scrape_instruction(
                        &instruction,
                        &correlation_id,
                        call.caller_id.clone(),
                        call.skill_name.clone(),
                    )
                    .await;
                #[allow(clippy::cast_possible_truncation)]
                let duration_ms = start.elapsed().as_millis() as u64;
                self.run_with_audit(
                    "web_scrape",
                    "web-scrape",
                    &redact_url_for_log(&instruction.url),
                    call.caller_id.clone(),
                    call.skill_name.clone(),
                    correlation_id,
                    duration_ms,
                    result,
                )
                .await
            }
            "fetch" => {
                let p: FetchParams = deserialize_params(&call.params)?;
                let correlation_id = EgressEvent::new_correlation_id();
                let start = Instant::now();
                let result = self
                    .handle_fetch(
                        &p,
                        &correlation_id,
                        call.caller_id.clone(),
                        call.skill_name.clone(),
                    )
                    .await;
                #[allow(clippy::cast_possible_truncation)]
                let duration_ms = start.elapsed().as_millis() as u64;
                self.run_with_audit(
                    "fetch",
                    "fetch",
                    &redact_url_for_log(&p.url),
                    call.caller_id.clone(),
                    call.skill_name.clone(),
                    correlation_id,
                    duration_ms,
                    result,
                )
                .await
            }
            _ => Ok(None),
        }
    }

    fn is_tool_retryable(&self, tool_id: &str) -> bool {
        matches!(tool_id, "web_scrape" | "fetch")
    }

    crate::tool_executor_no_inner_defaults!();
}

fn tool_error_to_audit_result(e: &ToolError) -> AuditResult {
    match e {
        ToolError::Blocked { command } => AuditResult::Blocked {
            reason: command.clone(),
        },
        ToolError::Timeout { .. } => AuditResult::Timeout,
        _ => AuditResult::Error {
            message: e.to_string(),
        },
    }
}

impl WebScrapeExecutor {
    #[allow(clippy::too_many_arguments)]
    async fn run_with_audit(
        &self,
        audit_tool_name: &str,
        public_tool_name: &str,
        audit_command: &str,
        caller_id: Option<String>,
        skill_name: Option<Vec<String>>,
        correlation_id: String,
        duration_ms: u64,
        result: Result<String, ToolError>,
    ) -> Result<Option<ToolOutput>, ToolError> {
        match result {
            Ok(output) => {
                self.log_audit(
                    audit_tool_name,
                    audit_command,
                    AuditResult::Success,
                    duration_ms,
                    None,
                    caller_id,
                    skill_name,
                    Some(correlation_id),
                )
                .await;
                Ok(Some(ToolOutput {
                    tool_name: ToolName::new(public_tool_name),
                    summary: output,
                    blocks_executed: 1,
                    filter_stats: None,
                    diff: None,
                    streamed: false,
                    terminal_id: None,
                    locations: None,
                    raw_response: None,
                    claim_source: Some(ClaimSource::WebScrape),
                    ..Default::default()
                }))
            }
            Err(e) => {
                let audit_result = tool_error_to_audit_result(&e);
                self.log_audit(
                    audit_tool_name,
                    audit_command,
                    audit_result,
                    duration_ms,
                    Some(&e),
                    caller_id,
                    skill_name,
                    Some(correlation_id),
                )
                .await;
                Err(e)
            }
        }
    }

    #[allow(clippy::too_many_arguments)] // function with many required inputs; a *Params struct would be more verbose without simplifying the call site
    async fn log_audit(
        &self,
        tool: &str,
        command: &str,
        result: AuditResult,
        duration_ms: u64,
        error: Option<&ToolError>,
        caller_id: Option<String>,
        skill_name: Option<Vec<String>>,
        correlation_id: Option<String>,
    ) {
        if let Some(ref logger) = self.audit_logger {
            let (error_category, error_domain, error_phase) =
                error.map_or((None, None, None), |e| {
                    let cat = e.category();
                    (
                        Some(cat.label().to_owned()),
                        Some(cat.domain().label().to_owned()),
                        Some(cat.phase().label().to_owned()),
                    )
                });
            let entry = AuditEntry {
                source_kind: None,
                trust_level: None,
                timestamp: chrono_now(),
                tool: tool.into(),
                command: command.into(),
                result,
                duration_ms,
                error_category,
                error_domain,
                error_phase,
                claim_source: Some(ClaimSource::WebScrape),
                mcp_server_id: None,
                injection_flagged: false,
                embedding_anomalous: false,
                cross_boundary_mcp_to_acp: false,
                adversarial_policy_decision: None,
                exit_code: None,
                truncated: false,
                caller_id,
                skill_name,
                policy_match: None,
                correlation_id,
                vigil_risk: None,
                execution_env: None,
                resolved_cwd: None,
                scope_at_definition: None,
                scope_at_dispatch: None,
            };
            logger.log(&entry).await;
        }
    }

    fn send_egress_event(&self, event: EgressEvent) {
        if let Some(ref tx) = self.egress_tx {
            match tx.try_send(event) {
                Ok(()) => {}
                Err(tokio::sync::mpsc::error::TrySendError::Full(_)) => {
                    self.egress_dropped.fetch_add(1, Ordering::Relaxed);
                }
                Err(tokio::sync::mpsc::error::TrySendError::Closed(_)) => {
                    tracing::debug!("egress channel closed; executor continuing without telemetry");
                }
            }
        }
    }

    async fn log_egress_event(&self, event: &EgressEvent) {
        if let Some(ref logger) = self.audit_logger {
            logger.log_egress(event).await;
        }
        self.send_egress_event(event.clone());
    }

    async fn handle_fetch(
        &self,
        params: &FetchParams,
        correlation_id: &str,
        caller_id: Option<String>,
        skill_name: Option<Vec<String>>,
    ) -> Result<String, ToolError> {
        let parsed = validate_url(&params.url);
        let host_str = parsed
            .as_ref()
            .map(|u| u.host_str().unwrap_or("").to_owned())
            .unwrap_or_default();

        if let Err(ref _e) = parsed {
            if self.egress_config.enabled && self.egress_config.log_blocked {
                let event = Self::make_blocked_event(
                    "fetch",
                    &params.url,
                    &host_str,
                    correlation_id,
                    caller_id.clone(),
                    skill_name.clone(),
                    "scheme",
                );
                self.log_egress_event(&event).await;
            }
            return Err(parsed.unwrap_err());
        }
        let parsed = parsed.unwrap();

        if let Err(e) = check_domain_policy(
            parsed.host_str().unwrap_or(""),
            &self.allowed_domains,
            &self.denied_domains,
        ) {
            if self.egress_config.enabled && self.egress_config.log_blocked {
                let event = Self::make_blocked_event(
                    "fetch",
                    &params.url,
                    parsed.host_str().unwrap_or(""),
                    correlation_id,
                    caller_id.clone(),
                    skill_name.clone(),
                    "blocklist",
                );
                self.log_egress_event(&event).await;
            }
            return Err(e);
        }

        let (host, addrs) = match resolve_and_validate(&parsed).await {
            Ok(v) => v,
            Err(e) => {
                if self.egress_config.enabled && self.egress_config.log_blocked {
                    let event = Self::make_blocked_event(
                        "fetch",
                        &params.url,
                        parsed.host_str().unwrap_or(""),
                        correlation_id,
                        caller_id.clone(),
                        skill_name.clone(),
                        "ssrf",
                    );
                    self.log_egress_event(&event).await;
                }
                return Err(e);
            }
        };

        let body = self
            .fetch_html(
                &params.url,
                &host,
                &addrs,
                "fetch",
                correlation_id,
                caller_id,
                skill_name,
            )
            .await?;
        self.apply_ipi_filter(&body, &params.url).await
    }

    async fn scrape_instruction(
        &self,
        instruction: &ScrapeInstruction,
        correlation_id: &str,
        caller_id: Option<String>,
        skill_name: Option<Vec<String>>,
    ) -> Result<String, ToolError> {
        let parsed = validate_url(&instruction.url);
        let host_str = parsed
            .as_ref()
            .map(|u| u.host_str().unwrap_or("").to_owned())
            .unwrap_or_default();

        if let Err(ref _e) = parsed {
            if self.egress_config.enabled && self.egress_config.log_blocked {
                let event = Self::make_blocked_event(
                    "web_scrape",
                    &instruction.url,
                    &host_str,
                    correlation_id,
                    caller_id.clone(),
                    skill_name.clone(),
                    "scheme",
                );
                self.log_egress_event(&event).await;
            }
            return Err(parsed.unwrap_err());
        }
        let parsed = parsed.unwrap();

        if let Err(e) = check_domain_policy(
            parsed.host_str().unwrap_or(""),
            &self.allowed_domains,
            &self.denied_domains,
        ) {
            if self.egress_config.enabled && self.egress_config.log_blocked {
                let event = Self::make_blocked_event(
                    "web_scrape",
                    &instruction.url,
                    parsed.host_str().unwrap_or(""),
                    correlation_id,
                    caller_id.clone(),
                    skill_name.clone(),
                    "blocklist",
                );
                self.log_egress_event(&event).await;
            }
            return Err(e);
        }

        let (host, addrs) = match resolve_and_validate(&parsed).await {
            Ok(v) => v,
            Err(e) => {
                if self.egress_config.enabled && self.egress_config.log_blocked {
                    let event = Self::make_blocked_event(
                        "web_scrape",
                        &instruction.url,
                        parsed.host_str().unwrap_or(""),
                        correlation_id,
                        caller_id.clone(),
                        skill_name.clone(),
                        "ssrf",
                    );
                    self.log_egress_event(&event).await;
                }
                return Err(e);
            }
        };

        let html = self
            .fetch_html(
                &instruction.url,
                &host,
                &addrs,
                "web_scrape",
                correlation_id,
                caller_id,
                skill_name,
            )
            .await?;
        let selector = instruction.select.clone();
        let extract = ExtractMode::parse(&instruction.extract);
        let limit = instruction.limit.unwrap_or(10);
        let extracted = tokio::task::spawn_blocking(move || {
            parse_and_extract(&html, &selector, &extract, limit)
        })
        .await
        .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))??;
        // apply_ipi_filter runs on plain extracted text, not raw HTML
        self.apply_ipi_filter(&extracted, &instruction.url).await
    }

    fn make_blocked_event(
        tool: &str,
        url: &str,
        host: &str,
        correlation_id: &str,
        caller_id: Option<String>,
        skill_name: Option<Vec<String>>,
        block_reason: &'static str,
    ) -> EgressEvent {
        EgressEvent {
            timestamp: chrono_now(),
            kind: "egress",
            correlation_id: correlation_id.to_owned(),
            tool: tool.into(),
            url: redact_url_for_log(url),
            host: host.to_owned(),
            method: "GET".to_owned(),
            status: None,
            duration_ms: 0,
            response_bytes: 0,
            blocked: true,
            block_reason: Some(block_reason),
            caller_id,
            skill_name,
            hop: 0,
        }
    }

    /// Fetches the HTML at `url`, manually following up to 3 redirects.
    ///
    /// Each redirect target is validated with `validate_url` and `resolve_and_validate`
    /// before following, preventing SSRF via redirect chains. When egress logging is
    /// enabled, one [`EgressEvent`] is emitted per hop.
    ///
    /// # Errors
    ///
    /// Returns `ToolError::Blocked` if any redirect target resolves to a private IP.
    /// Returns `ToolError::Execution` on HTTP errors, too-large bodies, or too many redirects.
    #[allow(clippy::too_many_lines, clippy::too_many_arguments)]
    async fn fetch_html(
        &self,
        url: &str,
        host: &str,
        addrs: &[SocketAddr],
        tool: &str,
        correlation_id: &str,
        caller_id: Option<String>,
        skill_name: Option<Vec<String>>,
    ) -> Result<String, ToolError> {
        const MAX_REDIRECTS: usize = 3;

        let mut current_url = url.to_owned();
        let mut current_host = host.to_owned();
        let mut current_addrs = addrs.to_vec();

        for hop in 0..=MAX_REDIRECTS {
            let hop_start = Instant::now();
            // Build a per-hop client pinned to the current hop's validated addresses.
            let client = self.build_client(&current_host, &current_addrs);
            let resp = client
                .get(&current_url)
                .send()
                .await
                .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())));

            let resp = match resp {
                Ok(r) => r,
                Err(e) => {
                    if self.egress_config.enabled {
                        #[allow(clippy::cast_possible_truncation)]
                        let duration_ms = hop_start.elapsed().as_millis() as u64;
                        let event = EgressEvent {
                            timestamp: chrono_now(),
                            kind: "egress",
                            correlation_id: correlation_id.to_owned(),
                            tool: tool.into(),
                            url: redact_url_for_log(&current_url),
                            host: current_host.clone(),
                            method: "GET".to_owned(),
                            status: None,
                            duration_ms,
                            response_bytes: 0,
                            blocked: false,
                            block_reason: None,
                            caller_id: caller_id.clone(),
                            skill_name: skill_name.clone(),
                            #[allow(clippy::cast_possible_truncation)]
                            hop: hop as u8,
                        };
                        self.log_egress_event(&event).await;
                    }
                    return Err(e);
                }
            };

            let status = resp.status();

            if status.is_redirection() {
                if hop == MAX_REDIRECTS {
                    return Err(ToolError::Execution(std::io::Error::other(
                        "too many redirects",
                    )));
                }

                let location = resp
                    .headers()
                    .get(reqwest::header::LOCATION)
                    .and_then(|v| v.to_str().ok())
                    .ok_or_else(|| {
                        ToolError::Execution(std::io::Error::other("redirect with no Location"))
                    })?;

                // Resolve relative redirect URLs against the current URL.
                let base = Url::parse(&current_url)
                    .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))?;
                let next_url = base
                    .join(location)
                    .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))?;

                let validated = validate_url(next_url.as_str());
                if let Err(ref _e) = validated {
                    if self.egress_config.enabled && self.egress_config.log_blocked {
                        #[allow(clippy::cast_possible_truncation)]
                        let duration_ms = hop_start.elapsed().as_millis() as u64;
                        let next_host = next_url.host_str().unwrap_or("").to_owned();
                        let event = EgressEvent {
                            timestamp: chrono_now(),
                            kind: "egress",
                            correlation_id: correlation_id.to_owned(),
                            tool: tool.into(),
                            url: redact_url_for_log(next_url.as_str()),
                            host: next_host,
                            method: "GET".to_owned(),
                            status: None,
                            duration_ms,
                            response_bytes: 0,
                            blocked: true,
                            block_reason: Some("ssrf"),
                            caller_id: caller_id.clone(),
                            skill_name: skill_name.clone(),
                            #[allow(clippy::cast_possible_truncation)]
                            hop: (hop + 1) as u8,
                        };
                        self.log_egress_event(&event).await;
                    }
                    return Err(validated.unwrap_err());
                }
                let validated = validated.unwrap();
                let resolve_result = resolve_and_validate(&validated).await;
                if let Err(ref _e) = resolve_result {
                    if self.egress_config.enabled && self.egress_config.log_blocked {
                        #[allow(clippy::cast_possible_truncation)]
                        let duration_ms = hop_start.elapsed().as_millis() as u64;
                        let next_host = next_url.host_str().unwrap_or("").to_owned();
                        let event = EgressEvent {
                            timestamp: chrono_now(),
                            kind: "egress",
                            correlation_id: correlation_id.to_owned(),
                            tool: tool.into(),
                            url: redact_url_for_log(next_url.as_str()),
                            host: next_host,
                            method: "GET".to_owned(),
                            status: None,
                            duration_ms,
                            response_bytes: 0,
                            blocked: true,
                            block_reason: Some("ssrf"),
                            caller_id: caller_id.clone(),
                            skill_name: skill_name.clone(),
                            #[allow(clippy::cast_possible_truncation)]
                            hop: (hop + 1) as u8,
                        };
                        self.log_egress_event(&event).await;
                    }
                    return Err(resolve_result.unwrap_err());
                }
                let (next_host, next_addrs) = resolve_result.unwrap();

                current_url = next_url.to_string();
                current_host = next_host;
                current_addrs = next_addrs;
                continue;
            }

            if !status.is_success() {
                if self.egress_config.enabled {
                    #[allow(clippy::cast_possible_truncation)]
                    let duration_ms = hop_start.elapsed().as_millis() as u64;
                    let event = EgressEvent {
                        timestamp: chrono_now(),
                        kind: "egress",
                        correlation_id: correlation_id.to_owned(),
                        tool: tool.into(),
                        url: redact_url_for_log(&current_url),
                        host: current_host.clone(),
                        method: "GET".to_owned(),
                        status: Some(status.as_u16()),
                        duration_ms,
                        response_bytes: 0,
                        blocked: false,
                        block_reason: None,
                        caller_id: caller_id.clone(),
                        skill_name: skill_name.clone(),
                        #[allow(clippy::cast_possible_truncation)]
                        hop: hop as u8,
                    };
                    self.log_egress_event(&event).await;
                }
                return Err(ToolError::Http {
                    status: status.as_u16(),
                    message: status.canonical_reason().unwrap_or("unknown").to_owned(),
                });
            }

            let bytes = resp
                .bytes()
                .await
                .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))?;

            if bytes.len() > self.max_body_bytes {
                if self.egress_config.enabled {
                    #[allow(clippy::cast_possible_truncation)]
                    let duration_ms = hop_start.elapsed().as_millis() as u64;
                    let event = EgressEvent {
                        timestamp: chrono_now(),
                        kind: "egress",
                        correlation_id: correlation_id.to_owned(),
                        tool: tool.into(),
                        url: redact_url_for_log(&current_url),
                        host: current_host.clone(),
                        method: "GET".to_owned(),
                        status: Some(status.as_u16()),
                        duration_ms,
                        response_bytes: bytes.len(),
                        blocked: false,
                        block_reason: None,
                        caller_id: caller_id.clone(),
                        skill_name: skill_name.clone(),
                        #[allow(clippy::cast_possible_truncation)]
                        hop: hop as u8,
                    };
                    self.log_egress_event(&event).await;
                }
                return Err(ToolError::Execution(std::io::Error::other(format!(
                    "response too large: {} bytes (max: {})",
                    bytes.len(),
                    self.max_body_bytes,
                ))));
            }

            // Success — emit egress event.
            if self.egress_config.enabled {
                #[allow(clippy::cast_possible_truncation)]
                let duration_ms = hop_start.elapsed().as_millis() as u64;
                let response_bytes = if self.egress_config.log_response_bytes {
                    bytes.len()
                } else {
                    0
                };
                let event = EgressEvent {
                    timestamp: chrono_now(),
                    kind: "egress",
                    correlation_id: correlation_id.to_owned(),
                    tool: tool.into(),
                    url: redact_url_for_log(&current_url),
                    host: current_host.clone(),
                    method: "GET".to_owned(),
                    status: Some(status.as_u16()),
                    duration_ms,
                    response_bytes,
                    blocked: false,
                    block_reason: None,
                    caller_id: caller_id.clone(),
                    skill_name: skill_name.clone(),
                    #[allow(clippy::cast_possible_truncation)]
                    hop: hop as u8,
                };
                self.log_egress_event(&event).await;
            }

            return String::from_utf8(bytes.to_vec())
                .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())));
        }

        Err(ToolError::Execution(std::io::Error::other(
            "too many redirects",
        )))
    }

    /// Apply IPI filter to a fetched response body.
    ///
    /// Runs regex scanning on a blocking thread via `spawn_blocking` to avoid
    /// stalling the tokio executor on large inputs. When score >= threshold,
    /// prepends a warning header and emits a `tracing::warn!` log.
    ///
    /// # Errors
    ///
    /// Returns a [`ToolError`] if the blocking scan task panics.
    #[tracing::instrument(name = "tools.scrape.apply_ipi_filter", skip(self, body), fields(body_len = body.len()))]
    async fn apply_ipi_filter(&self, body: &str, url: &str) -> Result<String, ToolError> {
        let verdict = self
            .ipi_filter
            .filter_async(body.to_owned())
            .await
            .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))?;
        if !verdict.patterns_found.is_empty() {
            tracing::warn!(
                url = url,
                score = verdict.score,
                patterns = ?verdict.patterns_found,
                "IPI patterns detected in fetched content"
            );
        }
        // verdict.sanitized == body only when score < threshold (no redaction)
        if verdict.sanitized == body {
            Ok(verdict.sanitized)
        } else {
            Ok(format!(
                "[IPI WARNING: score={:.2}, patterns={}] {}",
                verdict.score,
                verdict.patterns_found.join(", "),
                verdict.sanitized,
            ))
        }
    }
}

fn extract_scrape_blocks(text: &str) -> Vec<&str> {
    crate::executor::extract_fenced_blocks(text, "scrape")
}

/// Resolves DNS for the URL host, validates all resolved IPs against private ranges,
/// and returns the hostname and validated socket addresses.
///
/// Returning the addresses allows the caller to pin the HTTP client to these exact
/// addresses, eliminating TOCTOU between DNS validation and the actual connection.
///
/// Delegates the actual lookup+validation loop to the shared
/// [`zeph_common::net::resolve_and_validate`] helper (also used by `zeph-a2a`'s client)
/// and maps its neutral error into this crate's [`ToolError`].
#[tracing::instrument(name = "tools.scrape.dns.resolve", skip(url), fields(host = url.host_str().unwrap_or("")))]
async fn resolve_and_validate(url: &Url) -> Result<(String, Vec<SocketAddr>), ToolError> {
    let Some(host) = url.host_str() else {
        return Ok((String::new(), vec![]));
    };
    let port = url.port_or_known_default().unwrap_or(443);
    let addrs = zeph_common::net::resolve_and_validate(host, port)
        .await
        .map_err(|e| match e {
            zeph_common::net::ResolveError::Timeout(timeout) => ToolError::Timeout {
                timeout_secs: timeout.as_secs(),
            },
            zeph_common::net::ResolveError::Lookup(io_err) => ToolError::Blocked {
                command: format!("DNS resolution failed: {io_err}"),
            },
            zeph_common::net::ResolveError::PrivateAddress { host, addr } => ToolError::Blocked {
                command: format!("SSRF protection: private IP {addr} for host {host}"),
            },
            other => ToolError::Blocked {
                command: format!("DNS resolution failed: {other}"),
            },
        })?;
    Ok((host.to_owned(), addrs))
}

fn parse_and_extract(
    html: &str,
    selector: &str,
    extract: &ExtractMode,
    limit: usize,
) -> Result<String, ToolError> {
    let soup = scrape_core::Soup::parse(html);

    let tags = soup.find_all(selector).map_err(|e| {
        ToolError::Execution(std::io::Error::new(
            std::io::ErrorKind::InvalidData,
            format!("invalid selector: {e}"),
        ))
    })?;

    let mut results = Vec::new();

    for tag in tags.into_iter().take(limit) {
        let value = match extract {
            ExtractMode::Text => tag.text(),
            ExtractMode::Html => tag.inner_html(),
            ExtractMode::Attr(name) => tag.get(name).unwrap_or_default().to_owned(),
        };
        if !value.trim().is_empty() {
            results.push(value.trim().to_owned());
        }
    }

    if results.is_empty() {
        Ok(format!("No results for selector: {selector}"))
    } else {
        Ok(results.join("\n"))
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::domain_match::domain_matches;
    use crate::net::is_private_host;
    use std::assert_matches;

    // --- extract_scrape_blocks ---

    #[test]
    fn extract_single_block() {
        let text =
            "Here:\n```scrape\n{\"url\":\"https://example.com\",\"select\":\"h1\"}\n```\nDone.";
        let blocks = extract_scrape_blocks(text);
        assert_eq!(blocks.len(), 1);
        assert!(blocks[0].contains("example.com"));
    }

    #[test]
    fn extract_multiple_blocks() {
        let text = "```scrape\n{\"url\":\"https://a.com\",\"select\":\"h1\"}\n```\ntext\n```scrape\n{\"url\":\"https://b.com\",\"select\":\"p\"}\n```";
        let blocks = extract_scrape_blocks(text);
        assert_eq!(blocks.len(), 2);
    }

    #[test]
    fn no_blocks_returns_empty() {
        let blocks = extract_scrape_blocks("plain text, no code blocks");
        assert!(blocks.is_empty());
    }

    #[test]
    fn unclosed_block_ignored() {
        let blocks = extract_scrape_blocks("```scrape\n{\"url\":\"https://x.com\"}");
        assert!(blocks.is_empty());
    }

    #[test]
    fn non_scrape_block_ignored() {
        let text =
            "```bash\necho hi\n```\n```scrape\n{\"url\":\"https://x.com\",\"select\":\"h1\"}\n```";
        let blocks = extract_scrape_blocks(text);
        assert_eq!(blocks.len(), 1);
        assert!(blocks[0].contains("x.com"));
    }

    #[test]
    fn multiline_json_block() {
        let text =
            "```scrape\n{\n  \"url\": \"https://example.com\",\n  \"select\": \"h1\"\n}\n```";
        let blocks = extract_scrape_blocks(text);
        assert_eq!(blocks.len(), 1);
        let instr: ScrapeInstruction = serde_json::from_str(blocks[0]).unwrap();
        assert_eq!(instr.url, "https://example.com");
    }

    // --- ScrapeInstruction parsing ---

    #[test]
    fn parse_valid_instruction() {
        let json = r#"{"url":"https://example.com","select":"h1","extract":"text","limit":5}"#;
        let instr: ScrapeInstruction = serde_json::from_str(json).unwrap();
        assert_eq!(instr.url, "https://example.com");
        assert_eq!(instr.select, "h1");
        assert_eq!(instr.extract, "text");
        assert_eq!(instr.limit, Some(5));
    }

    #[test]
    fn parse_minimal_instruction() {
        let json = r#"{"url":"https://example.com","select":"p"}"#;
        let instr: ScrapeInstruction = serde_json::from_str(json).unwrap();
        assert_eq!(instr.extract, "text");
        assert!(instr.limit.is_none());
    }

    #[test]
    fn parse_attr_extract() {
        let json = r#"{"url":"https://example.com","select":"a","extract":"attr:href"}"#;
        let instr: ScrapeInstruction = serde_json::from_str(json).unwrap();
        assert_eq!(instr.extract, "attr:href");
    }

    #[test]
    fn parse_invalid_json_errors() {
        let result = serde_json::from_str::<ScrapeInstruction>("not json");
        assert!(result.is_err());
    }

    // --- ExtractMode ---

    #[test]
    fn extract_mode_text() {
        assert_matches!(ExtractMode::parse("text"), ExtractMode::Text);
    }

    #[test]
    fn extract_mode_html() {
        assert_matches!(ExtractMode::parse("html"), ExtractMode::Html);
    }

    #[test]
    fn extract_mode_attr() {
        let mode = ExtractMode::parse("attr:href");
        assert_matches!(mode, ExtractMode::Attr(ref s) if s == "href");
    }

    #[test]
    fn extract_mode_unknown_defaults_to_text() {
        assert_matches!(ExtractMode::parse("unknown"), ExtractMode::Text);
    }

    // validate_url is now shared in `crate::net` — see net.rs tests for scheme/private-host
    // coverage. `ftp_rejected`/`file_rejected`/`ipv6_loopback_blocked` etc. moved there.

    // --- parse_and_extract ---

    #[test]
    fn extract_text_from_html() {
        let html = "<html><body><h1>Hello World</h1><p>Content</p></body></html>";
        let result = parse_and_extract(html, "h1", &ExtractMode::Text, 10).unwrap();
        assert_eq!(result, "Hello World");
    }

    #[test]
    fn extract_multiple_elements() {
        let html = "<ul><li>A</li><li>B</li><li>C</li></ul>";
        let result = parse_and_extract(html, "li", &ExtractMode::Text, 10).unwrap();
        assert_eq!(result, "A\nB\nC");
    }

    #[test]
    fn extract_with_limit() {
        let html = "<ul><li>A</li><li>B</li><li>C</li></ul>";
        let result = parse_and_extract(html, "li", &ExtractMode::Text, 2).unwrap();
        assert_eq!(result, "A\nB");
    }

    #[test]
    fn extract_attr_href() {
        let html = r#"<a href="https://example.com">Link</a>"#;
        let result =
            parse_and_extract(html, "a", &ExtractMode::Attr("href".to_owned()), 10).unwrap();
        assert_eq!(result, "https://example.com");
    }

    #[test]
    fn extract_inner_html() {
        let html = "<div><span>inner</span></div>";
        let result = parse_and_extract(html, "div", &ExtractMode::Html, 10).unwrap();
        assert!(result.contains("<span>inner</span>"));
    }

    #[test]
    fn no_matches_returns_message() {
        let html = "<html><body><p>text</p></body></html>";
        let result = parse_and_extract(html, "h1", &ExtractMode::Text, 10).unwrap();
        assert!(result.starts_with("No results for selector:"));
    }

    #[test]
    fn empty_text_skipped() {
        let html = "<ul><li>  </li><li>A</li></ul>";
        let result = parse_and_extract(html, "li", &ExtractMode::Text, 10).unwrap();
        assert_eq!(result, "A");
    }

    #[test]
    fn invalid_selector_errors() {
        let html = "<html><body></body></html>";
        let result = parse_and_extract(html, "[[[invalid", &ExtractMode::Text, 10);
        assert!(result.is_err());
    }

    #[test]
    fn empty_html_returns_no_results() {
        let result = parse_and_extract("", "h1", &ExtractMode::Text, 10).unwrap();
        assert!(result.starts_with("No results for selector:"));
    }

    #[test]
    fn nested_selector() {
        let html = "<div><span>inner</span></div><span>outer</span>";
        let result = parse_and_extract(html, "div > span", &ExtractMode::Text, 10).unwrap();
        assert_eq!(result, "inner");
    }

    #[test]
    fn attr_missing_returns_empty() {
        let html = r"<a>No href</a>";
        let result =
            parse_and_extract(html, "a", &ExtractMode::Attr("href".to_owned()), 10).unwrap();
        assert!(result.starts_with("No results for selector:"));
    }

    #[test]
    fn extract_html_mode() {
        let html = "<div><b>bold</b> text</div>";
        let result = parse_and_extract(html, "div", &ExtractMode::Html, 10).unwrap();
        assert!(result.contains("<b>bold</b>"));
    }

    #[test]
    fn limit_zero_returns_no_results() {
        let html = "<ul><li>A</li><li>B</li></ul>";
        let result = parse_and_extract(html, "li", &ExtractMode::Text, 0).unwrap();
        assert!(result.starts_with("No results for selector:"));
    }

    // --- WebScrapeExecutor (no-network) ---

    #[tokio::test]
    async fn executor_no_blocks_returns_none() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let result = executor.execute("plain text").await;
        assert!(result.unwrap().is_none());
    }

    #[tokio::test]
    async fn executor_invalid_json_errors() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let response = "```scrape\nnot json\n```";
        let result = executor.execute(response).await;
        assert_matches!(result, Err(ToolError::Execution(_)));
    }

    #[tokio::test]
    async fn executor_blocked_url_errors() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let response = "```scrape\n{\"url\":\"http://example.com\",\"select\":\"h1\"}\n```";
        let result = executor.execute(response).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));
    }

    #[tokio::test]
    async fn executor_private_ip_blocked() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let response = "```scrape\n{\"url\":\"https://192.168.1.1/api\",\"select\":\"h1\"}\n```";
        let result = executor.execute(response).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));
    }

    #[tokio::test]
    async fn executor_unreachable_host_returns_error() {
        let config = ScrapeConfig {
            timeout: 1,
            max_body_bytes: 1_048_576,
            ..Default::default()
        };
        let executor = WebScrapeExecutor::new(&config);
        let response = "```scrape\n{\"url\":\"https://192.0.2.1:1/page\",\"select\":\"h1\"}\n```";
        let result = executor.execute(response).await;
        assert_matches!(result, Err(ToolError::Execution(_)));
    }

    #[tokio::test]
    async fn executor_localhost_url_blocked() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let response = "```scrape\n{\"url\":\"https://localhost:9999/api\",\"select\":\"h1\"}\n```";
        let result = executor.execute(response).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));
    }

    #[tokio::test]
    async fn executor_empty_text_returns_none() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let result = executor.execute("").await;
        assert!(result.unwrap().is_none());
    }

    #[tokio::test]
    async fn executor_multiple_blocks_first_blocked() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let response = "```scrape\n{\"url\":\"http://evil.com\",\"select\":\"h1\"}\n```\n\
             ```scrape\n{\"url\":\"https://ok.com\",\"select\":\"h1\"}\n```";
        let result = executor.execute(response).await;
        assert!(result.is_err());
    }

    #[test]
    fn validate_url_empty_string() {
        let err = validate_url("").unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    #[test]
    fn validate_url_javascript_scheme_blocked() {
        let err = validate_url("javascript:alert(1)").unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    #[test]
    fn validate_url_data_scheme_blocked() {
        let err = validate_url("data:text/html,<h1>hi</h1>").unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    #[test]
    fn is_private_host_public_domain_is_false() {
        let host: url::Host<&str> = url::Host::Domain("example.com");
        assert!(!is_private_host(&host));
    }

    #[test]
    fn is_private_host_localhost_is_true() {
        let host: url::Host<&str> = url::Host::Domain("localhost");
        assert!(is_private_host(&host));
    }

    #[test]
    fn is_private_host_ipv6_unspecified_is_true() {
        let host = url::Host::Ipv6(std::net::Ipv6Addr::UNSPECIFIED);
        assert!(is_private_host(&host));
    }

    #[test]
    fn is_private_host_public_ipv6_is_false() {
        let host = url::Host::Ipv6("2001:db8::1".parse().unwrap());
        assert!(!is_private_host(&host));
    }

    // --- fetch_html redirect logic: wiremock HTTP server tests ---
    //
    // These tests use a local wiremock server to exercise the redirect-following logic
    // in `fetch_html` without requiring an external HTTPS connection. The server binds to
    // 127.0.0.1, and tests call `fetch_html` directly (bypassing `validate_url`) to avoid
    // the SSRF guard that would otherwise block loopback connections.

    /// Helper: returns executor + (`server_url`, `server_addr`) from a running wiremock mock server.
    /// The server address is passed to `fetch_html` via `resolve_to_addrs` so the client
    /// connects to the mock instead of doing a real DNS lookup.
    async fn mock_server_executor() -> (WebScrapeExecutor, wiremock::MockServer) {
        let server = wiremock::MockServer::start().await;
        let executor = WebScrapeExecutor {
            timeout: Duration::from_secs(5),
            max_body_bytes: 1_048_576,
            allowed_domains: vec![],
            denied_domains: vec![],
            audit_logger: None,
            egress_config: EgressConfig::default(),
            egress_tx: None,
            egress_dropped: Arc::new(AtomicU64::new(0)),
            ipi_filter: IpiFilter::new(0.6),
        };
        (executor, server)
    }

    /// Parses the mock server's URI into (`host_str`, `socket_addr`) for use with `build_client`.
    fn server_host_and_addr(server: &wiremock::MockServer) -> (String, Vec<std::net::SocketAddr>) {
        let uri = server.uri();
        let url = Url::parse(&uri).unwrap();
        let host = url.host_str().unwrap_or("127.0.0.1").to_owned();
        let port = url.port().unwrap_or(80);
        let addr: std::net::SocketAddr = format!("{host}:{port}").parse().unwrap();
        (host, vec![addr])
    }

    /// Test-only redirect follower that mimics `fetch_html`'s loop but skips `validate_url` /
    /// `resolve_and_validate`. This lets us exercise the redirect-counting and
    /// missing-Location logic against a plain HTTP wiremock server.
    async fn follow_redirects_raw(
        executor: &WebScrapeExecutor,
        start_url: &str,
        host: &str,
        addrs: &[std::net::SocketAddr],
    ) -> Result<String, ToolError> {
        const MAX_REDIRECTS: usize = 3;
        let mut current_url = start_url.to_owned();
        let mut current_host = host.to_owned();
        let mut current_addrs = addrs.to_vec();

        for hop in 0..=MAX_REDIRECTS {
            let client = executor.build_client(&current_host, &current_addrs);
            let resp = client
                .get(&current_url)
                .send()
                .await
                .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))?;

            let status = resp.status();

            if status.is_redirection() {
                if hop == MAX_REDIRECTS {
                    return Err(ToolError::Execution(std::io::Error::other(
                        "too many redirects",
                    )));
                }

                let location = resp
                    .headers()
                    .get(reqwest::header::LOCATION)
                    .and_then(|v| v.to_str().ok())
                    .ok_or_else(|| {
                        ToolError::Execution(std::io::Error::other("redirect with no Location"))
                    })?;

                let base = Url::parse(&current_url)
                    .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))?;
                let next_url = base
                    .join(location)
                    .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))?;

                // Re-use same host/addrs (mock server is always the same endpoint).
                current_url = next_url.to_string();
                // Preserve host/addrs as-is since the mock server doesn't change.
                let _ = &mut current_host;
                let _ = &mut current_addrs;
                continue;
            }

            if !status.is_success() {
                return Err(ToolError::Execution(std::io::Error::other(format!(
                    "HTTP {status}",
                ))));
            }

            let bytes = resp
                .bytes()
                .await
                .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())))?;

            if bytes.len() > executor.max_body_bytes {
                return Err(ToolError::Execution(std::io::Error::other(format!(
                    "response too large: {} bytes (max: {})",
                    bytes.len(),
                    executor.max_body_bytes,
                ))));
            }

            return String::from_utf8(bytes.to_vec())
                .map_err(|e| ToolError::Execution(std::io::Error::other(e.to_string())));
        }

        Err(ToolError::Execution(std::io::Error::other(
            "too many redirects",
        )))
    }

    #[tokio::test]
    async fn fetch_html_success_returns_body() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        Mock::given(method("GET"))
            .and(path("/page"))
            .respond_with(ResponseTemplate::new(200).set_body_string("<h1>OK</h1>"))
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/page", server.uri());
        let result = executor
            .fetch_html(&url, &host, &addrs, "fetch", "test-cid", None, None)
            .await;
        assert!(result.is_ok(), "expected Ok, got: {result:?}");
        assert_eq!(result.unwrap(), "<h1>OK</h1>");
    }

    #[tokio::test]
    async fn fetch_html_non_2xx_returns_error() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        Mock::given(method("GET"))
            .and(path("/forbidden"))
            .respond_with(ResponseTemplate::new(403))
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/forbidden", server.uri());
        let result = executor
            .fetch_html(&url, &host, &addrs, "fetch", "test-cid", None, None)
            .await;
        assert!(result.is_err());
        let msg = result.unwrap_err().to_string();
        assert!(msg.contains("403"), "expected 403 in error: {msg}");
    }

    #[tokio::test]
    async fn fetch_html_404_returns_error() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        Mock::given(method("GET"))
            .and(path("/missing"))
            .respond_with(ResponseTemplate::new(404))
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/missing", server.uri());
        let result = executor
            .fetch_html(&url, &host, &addrs, "fetch", "test-cid", None, None)
            .await;
        assert!(result.is_err());
        let msg = result.unwrap_err().to_string();
        assert!(msg.contains("404"), "expected 404 in error: {msg}");
    }

    #[tokio::test]
    async fn fetch_html_redirect_no_location_returns_error() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        // 302 with no Location header
        Mock::given(method("GET"))
            .and(path("/redirect-no-loc"))
            .respond_with(ResponseTemplate::new(302))
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/redirect-no-loc", server.uri());
        let result = executor
            .fetch_html(&url, &host, &addrs, "fetch", "test-cid", None, None)
            .await;
        assert!(result.is_err());
        let msg = result.unwrap_err().to_string();
        assert!(
            msg.contains("Location") || msg.contains("location"),
            "expected Location-related error: {msg}"
        );
    }

    #[tokio::test]
    async fn fetch_html_single_redirect_followed() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        let final_url = format!("{}/final", server.uri());

        Mock::given(method("GET"))
            .and(path("/start"))
            .respond_with(ResponseTemplate::new(302).insert_header("location", final_url.as_str()))
            .mount(&server)
            .await;

        Mock::given(method("GET"))
            .and(path("/final"))
            .respond_with(ResponseTemplate::new(200).set_body_string("<p>final</p>"))
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/start", server.uri());
        let result = follow_redirects_raw(&executor, &url, &host, &addrs).await;
        assert!(result.is_ok(), "single redirect should succeed: {result:?}");
        assert_eq!(result.unwrap(), "<p>final</p>");
    }

    #[tokio::test]
    async fn fetch_html_three_redirects_allowed() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        let hop2 = format!("{}/hop2", server.uri());
        let hop3 = format!("{}/hop3", server.uri());
        let final_dest = format!("{}/done", server.uri());

        Mock::given(method("GET"))
            .and(path("/hop1"))
            .respond_with(ResponseTemplate::new(301).insert_header("location", hop2.as_str()))
            .mount(&server)
            .await;
        Mock::given(method("GET"))
            .and(path("/hop2"))
            .respond_with(ResponseTemplate::new(301).insert_header("location", hop3.as_str()))
            .mount(&server)
            .await;
        Mock::given(method("GET"))
            .and(path("/hop3"))
            .respond_with(ResponseTemplate::new(301).insert_header("location", final_dest.as_str()))
            .mount(&server)
            .await;
        Mock::given(method("GET"))
            .and(path("/done"))
            .respond_with(ResponseTemplate::new(200).set_body_string("<p>done</p>"))
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/hop1", server.uri());
        let result = follow_redirects_raw(&executor, &url, &host, &addrs).await;
        assert!(result.is_ok(), "3 redirects should succeed: {result:?}");
        assert_eq!(result.unwrap(), "<p>done</p>");
    }

    #[tokio::test]
    async fn fetch_html_four_redirects_rejected() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        let hop2 = format!("{}/r2", server.uri());
        let hop3 = format!("{}/r3", server.uri());
        let hop4 = format!("{}/r4", server.uri());
        let hop5 = format!("{}/r5", server.uri());

        for (from, to) in [
            ("/r1", &hop2),
            ("/r2", &hop3),
            ("/r3", &hop4),
            ("/r4", &hop5),
        ] {
            Mock::given(method("GET"))
                .and(path(from))
                .respond_with(ResponseTemplate::new(301).insert_header("location", to.as_str()))
                .mount(&server)
                .await;
        }

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/r1", server.uri());
        let result = follow_redirects_raw(&executor, &url, &host, &addrs).await;
        assert!(result.is_err(), "4 redirects should be rejected");
        let msg = result.unwrap_err().to_string();
        assert!(
            msg.contains("redirect"),
            "expected redirect-related error: {msg}"
        );
    }

    #[tokio::test]
    async fn fetch_html_body_too_large_returns_error() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let small_limit_executor = WebScrapeExecutor {
            timeout: Duration::from_secs(5),
            max_body_bytes: 10,
            allowed_domains: vec![],
            denied_domains: vec![],
            audit_logger: None,
            egress_config: EgressConfig::default(),
            egress_tx: None,
            egress_dropped: Arc::new(AtomicU64::new(0)),
            ipi_filter: IpiFilter::new(0.6),
        };
        let server = wiremock::MockServer::start().await;
        Mock::given(method("GET"))
            .and(path("/big"))
            .respond_with(
                ResponseTemplate::new(200)
                    .set_body_string("this body is definitely longer than ten bytes"),
            )
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/big", server.uri());
        let result = small_limit_executor
            .fetch_html(&url, &host, &addrs, "fetch", "test-cid", None, None)
            .await;
        assert!(result.is_err());
        let msg = result.unwrap_err().to_string();
        assert!(msg.contains("too large"), "expected too-large error: {msg}");
    }

    #[test]
    fn extract_scrape_blocks_empty_block_content() {
        let text = "```scrape\n\n```";
        let blocks = extract_scrape_blocks(text);
        assert_eq!(blocks.len(), 1);
        assert!(blocks[0].is_empty());
    }

    #[test]
    fn extract_scrape_blocks_whitespace_only() {
        let text = "```scrape\n   \n```";
        let blocks = extract_scrape_blocks(text);
        assert_eq!(blocks.len(), 1);
    }

    #[test]
    fn parse_and_extract_multiple_selectors() {
        let html = "<div><h1>Title</h1><p>Para</p></div>";
        let result = parse_and_extract(html, "h1, p", &ExtractMode::Text, 10).unwrap();
        assert!(result.contains("Title"));
        assert!(result.contains("Para"));
    }

    #[test]
    fn webscrape_executor_new_with_custom_config() {
        let config = ScrapeConfig {
            timeout: 60,
            max_body_bytes: 512,
            ..Default::default()
        };
        let executor = WebScrapeExecutor::new(&config);
        assert_eq!(executor.max_body_bytes, 512);
    }

    #[test]
    fn webscrape_executor_debug() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let dbg = format!("{executor:?}");
        assert!(dbg.contains("WebScrapeExecutor"));
    }

    #[test]
    fn extract_mode_attr_empty_name() {
        let mode = ExtractMode::parse("attr:");
        assert_matches!(mode, ExtractMode::Attr(ref s) if s.is_empty());
    }

    #[test]
    fn default_extract_returns_text() {
        assert_eq!(default_extract(), "text");
    }

    #[test]
    fn scrape_instruction_debug() {
        let json = r#"{"url":"https://example.com","select":"h1"}"#;
        let instr: ScrapeInstruction = serde_json::from_str(json).unwrap();
        let dbg = format!("{instr:?}");
        assert!(dbg.contains("ScrapeInstruction"));
    }

    #[test]
    fn extract_mode_debug() {
        let mode = ExtractMode::Text;
        let dbg = format!("{mode:?}");
        assert!(dbg.contains("Text"));
    }

    // --- fetch_html redirect logic: constant and validation unit tests ---

    /// `MAX_REDIRECTS` is 3; the 4th redirect attempt must be rejected.
    /// Verify the boundary is correct by inspecting the constant value.
    #[test]
    fn max_redirects_constant_is_three() {
        // fetch_html uses `for hop in 0..=MAX_REDIRECTS` and returns error when hop == MAX_REDIRECTS
        // while still in a redirect. That means hops 0,1,2 can redirect; hop 3 triggers the error.
        // This test documents the expected limit.
        const MAX_REDIRECTS: usize = 3;
        assert_eq!(MAX_REDIRECTS, 3, "fetch_html allows exactly 3 redirects");
    }

    /// Verifies that a Location-less redirect would produce an error string containing the
    /// expected message, matching the error path in `fetch_html`.
    #[test]
    fn redirect_no_location_error_message() {
        let err = std::io::Error::other("redirect with no Location");
        assert!(err.to_string().contains("redirect with no Location"));
    }

    /// Verifies that a too-many-redirects condition produces the expected error string.
    #[test]
    fn too_many_redirects_error_message() {
        let err = std::io::Error::other("too many redirects");
        assert!(err.to_string().contains("too many redirects"));
    }

    /// Verifies that a non-2xx HTTP status produces an error message with the status code.
    #[test]
    fn non_2xx_status_error_format() {
        let status = reqwest::StatusCode::FORBIDDEN;
        let msg = format!("HTTP {status}");
        assert!(msg.contains("403"));
    }

    /// Verifies that a 404 response status code formats into the expected error message.
    #[test]
    fn not_found_status_error_format() {
        let status = reqwest::StatusCode::NOT_FOUND;
        let msg = format!("HTTP {status}");
        assert!(msg.contains("404"));
    }

    /// Verifies relative redirect resolution for same-host paths (simulates Location: /other).
    #[test]
    fn relative_redirect_same_host_path() {
        let base = Url::parse("https://example.com/current").unwrap();
        let resolved = base.join("/other").unwrap();
        assert_eq!(resolved.as_str(), "https://example.com/other");
    }

    /// Verifies relative redirect resolution preserves scheme and host.
    #[test]
    fn relative_redirect_relative_path() {
        let base = Url::parse("https://example.com/a/b").unwrap();
        let resolved = base.join("c").unwrap();
        assert_eq!(resolved.as_str(), "https://example.com/a/c");
    }

    /// Verifies that an absolute redirect URL overrides base URL completely.
    #[test]
    fn absolute_redirect_overrides_base() {
        let base = Url::parse("https://example.com/page").unwrap();
        let resolved = base.join("https://other.com/target").unwrap();
        assert_eq!(resolved.as_str(), "https://other.com/target");
    }

    /// Verifies that a redirect Location of http:// (downgrade) is rejected.
    #[test]
    fn redirect_http_downgrade_rejected() {
        let location = "http://example.com/page";
        let base = Url::parse("https://example.com/start").unwrap();
        let next = base.join(location).unwrap();
        let err = validate_url(next.as_str()).unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    /// Verifies that a redirect to a private IP literal is blocked.
    #[test]
    fn redirect_location_private_ip_blocked() {
        let location = "https://192.168.100.1/admin";
        let base = Url::parse("https://example.com/start").unwrap();
        let next = base.join(location).unwrap();
        let err = validate_url(next.as_str()).unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
        let ToolError::Blocked { command: cmd } = err else {
            panic!("expected Blocked");
        };
        assert!(
            cmd.contains("private") || cmd.contains("scheme"),
            "error message should describe the block reason: {cmd}"
        );
    }

    /// Verifies that a redirect to a .internal domain is blocked.
    #[test]
    fn redirect_location_internal_domain_blocked() {
        let location = "https://metadata.internal/latest/meta-data/";
        let base = Url::parse("https://example.com/start").unwrap();
        let next = base.join(location).unwrap();
        let err = validate_url(next.as_str()).unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    /// Verifies that a chain of 3 valid public redirects passes `validate_url` at every hop.
    #[test]
    fn redirect_chain_three_hops_all_public() {
        let hops = [
            "https://redirect1.example.com/hop1",
            "https://redirect2.example.com/hop2",
            "https://destination.example.com/final",
        ];
        for hop in hops {
            assert!(validate_url(hop).is_ok(), "expected ok for {hop}");
        }
    }

    // --- SSRF redirect chain defense ---

    /// Verifies that a redirect Location pointing to a private IP is rejected by `validate_url`
    /// before any connection attempt — simulating the validation step inside `fetch_html`.
    #[test]
    fn redirect_to_private_ip_rejected_by_validate_url() {
        // These would appear as Location headers in a redirect response.
        let private_targets = [
            "https://127.0.0.1/secret",
            "https://10.0.0.1/internal",
            "https://192.168.1.1/admin",
            "https://172.16.0.1/data",
            "https://[::1]/path",
            "https://[fe80::1]/path",
            "https://localhost/path",
            "https://service.internal/api",
        ];
        for target in private_targets {
            let result = validate_url(target);
            assert!(result.is_err(), "expected error for {target}");
            assert!(
                matches!(result.unwrap_err(), ToolError::Blocked { .. }),
                "expected Blocked for {target}"
            );
        }
    }

    /// Verifies that relative redirect URLs are resolved correctly before validation.
    #[test]
    fn redirect_relative_url_resolves_correctly() {
        let base = Url::parse("https://example.com/page").unwrap();
        let relative = "/other";
        let resolved = base.join(relative).unwrap();
        assert_eq!(resolved.as_str(), "https://example.com/other");
    }

    /// Verifies that a protocol-relative redirect to http:// is rejected (scheme check).
    #[test]
    fn redirect_to_http_rejected() {
        let err = validate_url("http://example.com/page").unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    #[test]
    fn ipv4_mapped_ipv6_link_local_blocked() {
        let err = validate_url("https://[::ffff:169.254.0.1]/path").unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    #[test]
    fn ipv4_mapped_ipv6_public_allowed() {
        assert!(validate_url("https://[::ffff:93.184.216.34]/path").is_ok());
    }

    // --- fetch tool ---

    #[tokio::test]
    async fn fetch_http_scheme_blocked() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let call = crate::executor::ToolCall {
            tool_id: ToolName::new("fetch"),
            params: {
                let mut m = serde_json::Map::new();
                m.insert("url".to_owned(), serde_json::json!("http://example.com"));
                m
            },
            caller_id: None,
            context: None,

            tool_call_id: String::new(),
            skill_name: None,
        };
        let result = executor.execute_tool_call(&call).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));
    }

    #[tokio::test]
    async fn fetch_private_ip_blocked() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let call = crate::executor::ToolCall {
            tool_id: ToolName::new("fetch"),
            params: {
                let mut m = serde_json::Map::new();
                m.insert(
                    "url".to_owned(),
                    serde_json::json!("https://192.168.1.1/secret"),
                );
                m
            },
            caller_id: None,
            context: None,

            tool_call_id: String::new(),
            skill_name: None,
        };
        let result = executor.execute_tool_call(&call).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));
    }

    #[tokio::test]
    async fn fetch_localhost_blocked() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let call = crate::executor::ToolCall {
            tool_id: ToolName::new("fetch"),
            params: {
                let mut m = serde_json::Map::new();
                m.insert(
                    "url".to_owned(),
                    serde_json::json!("https://localhost/page"),
                );
                m
            },
            caller_id: None,
            context: None,

            tool_call_id: String::new(),
            skill_name: None,
        };
        let result = executor.execute_tool_call(&call).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));
    }

    #[tokio::test]
    async fn fetch_unknown_tool_returns_none() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let call = crate::executor::ToolCall {
            tool_id: ToolName::new("unknown_tool"),
            params: serde_json::Map::new(),
            caller_id: None,
            context: None,

            tool_call_id: String::new(),
            skill_name: None,
        };
        let result = executor.execute_tool_call(&call).await;
        assert!(result.unwrap().is_none());
    }

    #[tokio::test]
    async fn fetch_returns_body_via_mock() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        Mock::given(method("GET"))
            .and(path("/content"))
            .respond_with(ResponseTemplate::new(200).set_body_string("plain text content"))
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        let url = format!("{}/content", server.uri());
        let result = executor
            .fetch_html(&url, &host, &addrs, "fetch", "test-cid", None, None)
            .await;
        assert!(result.is_ok());
        assert_eq!(result.unwrap(), "plain text content");
    }

    #[test]
    fn tool_definitions_returns_web_scrape_and_fetch() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let defs = executor.tool_definitions();
        assert_eq!(defs.len(), 2);
        assert_eq!(defs[0].id, "web_scrape");
        assert_eq!(
            defs[0].invocation,
            crate::registry::InvocationHint::FencedBlock("scrape")
        );
        assert_eq!(defs[1].id, "fetch");
        assert_eq!(
            defs[1].invocation,
            crate::registry::InvocationHint::ToolCall
        );
    }

    #[test]
    fn tool_definitions_schema_has_all_params() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        let defs = executor.tool_definitions();
        let obj = defs[0].schema.as_object().unwrap();
        let props = obj["properties"].as_object().unwrap();
        assert!(props.contains_key("url"));
        assert!(props.contains_key("select"));
        assert!(props.contains_key("extract"));
        assert!(props.contains_key("limit"));
        let req = obj["required"].as_array().unwrap();
        assert!(req.iter().any(|v| v.as_str() == Some("url")));
        assert!(req.iter().any(|v| v.as_str() == Some("select")));
        assert!(!req.iter().any(|v| v.as_str() == Some("extract")));
    }

    // --- is_private_host: new domain checks (AUD-02) ---

    #[test]
    fn subdomain_localhost_blocked() {
        let host: url::Host<&str> = url::Host::Domain("foo.localhost");
        assert!(is_private_host(&host));
    }

    #[test]
    fn internal_tld_blocked() {
        let host: url::Host<&str> = url::Host::Domain("service.internal");
        assert!(is_private_host(&host));
    }

    #[test]
    fn local_tld_blocked() {
        let host: url::Host<&str> = url::Host::Domain("printer.local");
        assert!(is_private_host(&host));
    }

    #[test]
    fn public_domain_not_blocked() {
        let host: url::Host<&str> = url::Host::Domain("example.com");
        assert!(!is_private_host(&host));
    }

    // --- resolve_and_validate: private IP rejection ---

    #[tokio::test]
    async fn resolve_loopback_rejected() {
        // 127.0.0.1 resolves directly (literal IP in DNS query)
        let url = url::Url::parse("https://127.0.0.1/path").unwrap();
        // validate_url catches this before resolve_and_validate, but test directly
        let result = resolve_and_validate(&url).await;
        assert!(
            result.is_err(),
            "loopback IP must be rejected by resolve_and_validate"
        );
        let err = result.unwrap_err();
        assert_matches!(err, crate::executor::ToolError::Blocked { .. });
    }

    #[tokio::test]
    async fn resolve_private_10_rejected() {
        let url = url::Url::parse("https://10.0.0.1/path").unwrap();
        let result = resolve_and_validate(&url).await;
        assert!(result.is_err());
        assert_matches!(
            result.unwrap_err(),
            crate::executor::ToolError::Blocked { .. }
        );
    }

    #[tokio::test]
    async fn resolve_private_192_rejected() {
        let url = url::Url::parse("https://192.168.1.1/path").unwrap();
        let result = resolve_and_validate(&url).await;
        assert!(result.is_err());
        assert_matches!(
            result.unwrap_err(),
            crate::executor::ToolError::Blocked { .. }
        );
    }

    #[tokio::test]
    async fn resolve_ipv6_loopback_rejected() {
        let url = url::Url::parse("https://[::1]/path").unwrap();
        let result = resolve_and_validate(&url).await;
        assert!(result.is_err());
        assert_matches!(
            result.unwrap_err(),
            crate::executor::ToolError::Blocked { .. }
        );
    }

    #[tokio::test]
    async fn resolve_no_host_returns_ok() {
        // URL without a resolvable host — should pass through
        let url = url::Url::parse("https://example.com/path").unwrap();
        // We can't do a live DNS test, but we can verify a URL with no host
        let url_no_host = url::Url::parse("data:text/plain,hello").unwrap();
        // data: URLs have no host; resolve_and_validate should return Ok with empty addrs
        let result = resolve_and_validate(&url_no_host).await;
        assert!(result.is_ok());
        let (host, addrs) = result.unwrap();
        assert!(host.is_empty());
        assert!(addrs.is_empty());
        drop(url);
        drop(url_no_host);
    }

    // --- audit logging ---

    /// Helper: build an `AuditLogger` writing to a temp file, and return the logger + path.
    async fn make_file_audit_logger(
        dir: &tempfile::TempDir,
    ) -> (
        std::sync::Arc<crate::audit::AuditLogger>,
        std::path::PathBuf,
    ) {
        use crate::audit::AuditLogger;
        use crate::config::AuditConfig;
        let path = dir.path().join("audit.log");
        let config = AuditConfig {
            enabled: true,
            destination: crate::config::AuditDestination::File(path.clone()),
            ..Default::default()
        };
        let logger = std::sync::Arc::new(AuditLogger::from_config(&config, false).await.unwrap());
        (logger, path)
    }

    #[tokio::test]
    async fn with_audit_sets_logger() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        assert!(executor.audit_logger.is_none());

        let dir = tempfile::tempdir().unwrap();
        let (logger, _path) = make_file_audit_logger(&dir).await;
        let executor = executor.with_audit(logger);
        assert!(executor.audit_logger.is_some());
    }

    #[test]
    fn tool_error_to_audit_result_blocked_maps_correctly() {
        let err = ToolError::Blocked {
            command: "scheme not allowed: http".into(),
        };
        let result = tool_error_to_audit_result(&err);
        assert!(
            matches!(result, AuditResult::Blocked { reason } if reason == "scheme not allowed: http")
        );
    }

    #[test]
    fn tool_error_to_audit_result_timeout_maps_correctly() {
        let err = ToolError::Timeout { timeout_secs: 15 };
        let result = tool_error_to_audit_result(&err);
        assert_matches!(result, AuditResult::Timeout);
    }

    #[test]
    fn tool_error_to_audit_result_execution_error_maps_correctly() {
        let err = ToolError::Execution(std::io::Error::other("connection refused"));
        let result = tool_error_to_audit_result(&err);
        assert!(
            matches!(result, AuditResult::Error { message } if message.contains("connection refused"))
        );
    }

    #[tokio::test]
    async fn fetch_audit_blocked_url_logged() {
        let dir = tempfile::tempdir().unwrap();
        let (logger, log_path) = make_file_audit_logger(&dir).await;

        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config).with_audit(logger);

        let call = crate::executor::ToolCall {
            tool_id: ToolName::new("fetch"),
            params: {
                let mut m = serde_json::Map::new();
                m.insert("url".to_owned(), serde_json::json!("http://example.com"));
                m
            },
            caller_id: None,
            context: None,

            tool_call_id: String::new(),
            skill_name: None,
        };
        let result = executor.execute_tool_call(&call).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));

        let content = tokio::fs::read_to_string(&log_path).await.unwrap();
        assert!(
            content.contains("\"tool\":\"fetch\""),
            "expected tool=fetch in audit: {content}"
        );
        assert!(
            content.contains("\"type\":\"blocked\""),
            "expected type=blocked in audit: {content}"
        );
        assert!(
            content.contains("http://example.com"),
            "expected URL in audit command field: {content}"
        );
    }

    #[tokio::test]
    async fn log_audit_success_writes_to_file() {
        let dir = tempfile::tempdir().unwrap();
        let (logger, log_path) = make_file_audit_logger(&dir).await;

        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config).with_audit(logger);

        executor
            .log_audit(
                "fetch",
                "https://example.com/page",
                AuditResult::Success,
                42,
                None,
                None,
                None,
                None,
            )
            .await;

        let content = tokio::fs::read_to_string(&log_path).await.unwrap();
        assert!(
            content.contains("\"tool\":\"fetch\""),
            "expected tool=fetch in audit: {content}"
        );
        assert!(
            content.contains("\"type\":\"success\""),
            "expected type=success in audit: {content}"
        );
        assert!(
            content.contains("\"command\":\"https://example.com/page\""),
            "expected command URL in audit: {content}"
        );
        assert!(
            content.contains("\"duration_ms\":42"),
            "expected duration_ms in audit: {content}"
        );
    }

    #[tokio::test]
    async fn log_audit_blocked_writes_to_file() {
        let dir = tempfile::tempdir().unwrap();
        let (logger, log_path) = make_file_audit_logger(&dir).await;

        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config).with_audit(logger);

        executor
            .log_audit(
                "web_scrape",
                "http://evil.com/page",
                AuditResult::Blocked {
                    reason: "scheme not allowed: http".into(),
                },
                0,
                None,
                None,
                None,
                None,
            )
            .await;

        let content = tokio::fs::read_to_string(&log_path).await.unwrap();
        assert!(
            content.contains("\"tool\":\"web_scrape\""),
            "expected tool=web_scrape in audit: {content}"
        );
        assert!(
            content.contains("\"type\":\"blocked\""),
            "expected type=blocked in audit: {content}"
        );
        assert!(
            content.contains("scheme not allowed"),
            "expected block reason in audit: {content}"
        );
    }

    #[tokio::test]
    async fn web_scrape_audit_blocked_url_logged() {
        let dir = tempfile::tempdir().unwrap();
        let (logger, log_path) = make_file_audit_logger(&dir).await;

        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config).with_audit(logger);

        let call = crate::executor::ToolCall {
            tool_id: ToolName::new("web_scrape"),
            params: {
                let mut m = serde_json::Map::new();
                m.insert("url".to_owned(), serde_json::json!("http://example.com"));
                m.insert("select".to_owned(), serde_json::json!("h1"));
                m
            },
            caller_id: None,
            context: None,

            tool_call_id: String::new(),
            skill_name: None,
        };
        let result = executor.execute_tool_call(&call).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));

        let content = tokio::fs::read_to_string(&log_path).await.unwrap();
        assert!(
            content.contains("\"tool\":\"web_scrape\""),
            "expected tool=web_scrape in audit: {content}"
        );
        assert!(
            content.contains("\"type\":\"blocked\""),
            "expected type=blocked in audit: {content}"
        );
    }

    #[tokio::test]
    async fn no_audit_logger_does_not_panic_on_blocked_fetch() {
        let config = ScrapeConfig::default();
        let executor = WebScrapeExecutor::new(&config);
        assert!(executor.audit_logger.is_none());

        let call = crate::executor::ToolCall {
            tool_id: ToolName::new("fetch"),
            params: {
                let mut m = serde_json::Map::new();
                m.insert("url".to_owned(), serde_json::json!("http://example.com"));
                m
            },
            caller_id: None,
            context: None,

            tool_call_id: String::new(),
            skill_name: None,
        };
        // Must not panic even without an audit logger
        let result = executor.execute_tool_call(&call).await;
        assert_matches!(result, Err(ToolError::Blocked { .. }));
    }

    // CR-10: fetch end-to-end via execute_tool_call -> handle_fetch -> fetch_html
    #[tokio::test]
    async fn fetch_execute_tool_call_end_to_end() {
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, ResponseTemplate};

        let (executor, server) = mock_server_executor().await;
        Mock::given(method("GET"))
            .and(path("/e2e"))
            .respond_with(ResponseTemplate::new(200).set_body_string("<h1>end-to-end</h1>"))
            .mount(&server)
            .await;

        let (host, addrs) = server_host_and_addr(&server);
        // Call fetch_html directly (bypassing SSRF guard for loopback mock server)
        let result = executor
            .fetch_html(
                &format!("{}/e2e", server.uri()),
                &host,
                &addrs,
                "fetch",
                "test-cid",
                None,
                None,
            )
            .await;
        assert!(result.is_ok());
        assert!(result.unwrap().contains("end-to-end"));
    }

    // --- domain_matches ---

    #[test]
    fn domain_matches_exact() {
        assert!(domain_matches("example.com", "example.com"));
        assert!(!domain_matches("example.com", "other.com"));
        assert!(!domain_matches("example.com", "sub.example.com"));
    }

    #[test]
    fn domain_matches_wildcard_single_subdomain() {
        assert!(domain_matches("*.example.com", "sub.example.com"));
        assert!(!domain_matches("*.example.com", "example.com"));
        assert!(!domain_matches("*.example.com", "sub.sub.example.com"));
    }

    #[test]
    fn domain_matches_wildcard_does_not_match_empty_label() {
        // Pattern "*.example.com" requires a non-empty label before ".example.com"
        assert!(!domain_matches("*.example.com", ".example.com"));
    }

    #[test]
    fn domain_matches_multi_wildcard_treated_as_exact() {
        // Multiple wildcards are unsupported — treated as literal pattern
        assert!(!domain_matches("*.*.example.com", "a.b.example.com"));
    }

    // --- check_domain_policy ---

    #[test]
    fn check_domain_policy_empty_lists_allow_all() {
        assert!(check_domain_policy("example.com", &[], &[]).is_ok());
        assert!(check_domain_policy("evil.com", &[], &[]).is_ok());
    }

    #[test]
    fn check_domain_policy_denylist_blocks() {
        let denied = vec!["evil.com".to_string()];
        let err = check_domain_policy("evil.com", &[], &denied).unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    #[test]
    fn check_domain_policy_denylist_does_not_block_other_domains() {
        let denied = vec!["evil.com".to_string()];
        assert!(check_domain_policy("good.com", &[], &denied).is_ok());
    }

    #[test]
    fn check_domain_policy_allowlist_permits_matching() {
        let allowed = vec!["docs.rs".to_string(), "*.rust-lang.org".to_string()];
        assert!(check_domain_policy("docs.rs", &allowed, &[]).is_ok());
        assert!(check_domain_policy("blog.rust-lang.org", &allowed, &[]).is_ok());
    }

    #[test]
    fn check_domain_policy_allowlist_blocks_unknown() {
        let allowed = vec!["docs.rs".to_string()];
        let err = check_domain_policy("other.com", &allowed, &[]).unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    #[test]
    fn check_domain_policy_deny_overrides_allow() {
        let allowed = vec!["example.com".to_string()];
        let denied = vec!["example.com".to_string()];
        let err = check_domain_policy("example.com", &allowed, &denied).unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
    }

    #[test]
    fn check_domain_policy_wildcard_in_denylist() {
        let denied = vec!["*.evil.com".to_string()];
        let err = check_domain_policy("sub.evil.com", &[], &denied).unwrap_err();
        assert_matches!(err, ToolError::Blocked { .. });
        // parent domain not blocked
        assert!(check_domain_policy("evil.com", &[], &denied).is_ok());
    }
}