oxi_agent/tools/

grep.rs

use super::path_security::PathGuard;
/// Grep tool - search files for patterns
use super::{AgentTool, AgentToolResult, ToolContext, ToolError};
use crate::tools::typed::TypedTool;
use async_trait::async_trait;
use regex::RegexBuilder;
use schemars::JsonSchema;
use serde::Deserialize;
use serde_json::{Value, json};
use std::path::{Path, PathBuf};
use tokio::fs;
use tokio::sync::oneshot;

/// Maximum characters per line in grep output
const GREP_MAX_LINE_LENGTH: usize = 500;

/// Truncate a single line to max characters, adding "... [truncated]" suffix.
fn truncate_line(line: &str) -> (String, bool) {
    if line.len() <= GREP_MAX_LINE_LENGTH {
        (line.to_string(), false)
    } else {
        (
            format!("{}... [truncated]", &line[..GREP_MAX_LINE_LENGTH]),
            true,
        )
    }
}

/// Typed arguments for [`GrepTool`].
///
/// Deserialized from the LLM's JSON tool call parameters.
#[derive(Deserialize, JsonSchema)]
pub struct GrepArgs {
    pattern: String,
    #[serde(default)]
    path: String,
    #[serde(default)]
    case_insensitive: bool,
    #[serde(default)]
    literal: bool,
    #[serde(default)]
    context: usize,
    include: Option<String>,
    #[serde(default = "default_grep_results")]
    max_results: usize,
}

fn default_grep_results() -> usize {
    100
}

/// GrepTool.
pub struct GrepTool {
    root_dir: Option<PathBuf>,
}

impl GrepTool {
    /// Create with no explicit root (uses ToolContext.workspace_dir at runtime).
    pub fn new() -> Self {
        Self { root_dir: None }
    }

    /// Create with a specific working directory (overrides ToolContext).
    pub fn with_cwd(cwd: PathBuf) -> Self {
        Self {
            root_dir: Some(cwd),
        }
    }

    /// Check if a filename matches a simple glob pattern like "*.rs", "*.ts"
    fn matches_glob(file_name: &str, pattern: &str) -> bool {
        if let Some(ext) = pattern.strip_prefix("*.") {
            file_name.ends_with(ext)
        } else if pattern.contains('*') {
            // Simple wildcard matching
            let parts: Vec<&str> = pattern.split('*').collect();
            if parts.len() == 2 {
                file_name.starts_with(parts[0]) && file_name.ends_with(parts[1])
            } else {
                file_name == pattern
            }
        } else {
            file_name == pattern
        }
    }

    #[allow(clippy::type_complexity)]
    #[allow(clippy::too_many_arguments)]
    async fn grep_impl(
        root_dir: &Path,
        pattern: &str,
        path: &str,
        case_insensitive: bool,
        literal: bool,
        context_before: usize,
        context_after: usize,
        include: Option<&str>,
        max_results: usize,
    ) -> Result<(String, bool), ToolError> {
        // Security: validate path with PathGuard
        let guard = PathGuard::new(root_dir);
        let root = guard
            .validate_traversal(Path::new(path))
            .map_err(|e| e.to_string())?;

        if !root.exists() {
            return Err(format!("Path not found: {}", path));
        }

        // Escape the pattern for literal matching if needed
        let pattern = if literal {
            regex::escape(pattern)
        } else {
            pattern.to_string()
        };

        let re = RegexBuilder::new(&pattern)
            .case_insensitive(case_insensitive)
            .build()
            .map_err(|e| format!("Invalid pattern '{}': {}", pattern, e))?;

        let mut matches: Vec<String> = Vec::new();
        let mut lines_truncated = false;
        Self::grep_walk(
            &root,
            &root,
            &re,
            include,
            context_before,
            context_after,
            max_results,
            &mut matches,
            &mut lines_truncated,
        )
        .await?;

        if matches.is_empty() {
            Ok(("No matches found".to_string(), false))
        } else {
            let header = format!("Found {} matches:\n", matches.len());
            Ok((header + &matches.join("\n"), lines_truncated))
        }
    }

    /// Read a file and return lines as a vector
    async fn read_file_lines(path: &Path) -> Result<Vec<String>, ToolError> {
        match fs::read_to_string(path).await {
            Ok(content) => {
                // Normalize line endings: replace CRLF and standalone CR with LF, then split
                let normalized = content.replace("\r\n", "\n").replace('\r', "\n");
                Ok(normalized.lines().map(|s| s.to_string()).collect())
            }
            Err(e) => Err(format!("Cannot read file: {}", e)),
        }
    }

    #[allow(clippy::too_many_arguments)]
    async fn grep_walk(
        root: &Path,
        current: &Path,
        re: &regex::Regex,
        include: Option<&str>,
        context_before: usize,
        context_after: usize,
        max_results: usize,
        matches: &mut Vec<String>,
        lines_truncated: &mut bool,
    ) -> Result<(), ToolError> {
        if matches.len() >= max_results {
            return Ok(());
        }

        // Detect and skip broken symlinks - they cause read_dir to fail
        // and should not cause the entire search to fail
        if current
            .symlink_metadata()
            .map(|m| m.file_type().is_symlink())
            .unwrap_or(false)
            && !current.exists()
        {
            return Ok(());
        }

        if current.is_file() {
            // Check include filter
            if let Some(glob) = include {
                let file_name = current
                    .file_name()
                    .map(|n| n.to_string_lossy().to_string())
                    .unwrap_or_default();
                if !Self::matches_glob(&file_name, glob) {
                    return Ok(());
                }
            }

            // Try to read and search the file
            match Self::read_file_lines(current).await {
                Ok(lines) => {
                    let relative = current.strip_prefix(root).unwrap_or(current).display();

                    for (i, line) in lines.iter().enumerate() {
                        if re.is_match(line) {
                            // Check if adding this match would exceed max_results
                            // We may need to add context lines too
                            let context_lines_count = if context_before > 0 || context_after > 0 {
                                let start = if context_before > 0 {
                                    i.saturating_sub(context_before)
                                } else {
                                    i
                                };
                                let end = std::cmp::min(lines.len(), i + context_after + 1);
                                end - start
                            } else {
                                1
                            };

                            if matches.len() + context_lines_count > max_results {
                                // Can't add this match with its context, stop
                                return Ok(());
                            }

                            // Add context lines before match
                            if context_before > 0 && i > 0 {
                                let start = i.saturating_sub(context_before);
                                for (j, context_line) in
                                    lines.iter().enumerate().take(i).skip(start)
                                {
                                    let (truncated_text, was_truncated) =
                                        truncate_line(context_line);
                                    if was_truncated {
                                        *lines_truncated = true;
                                    }
                                    matches.push(format!(
                                        "{}-{}- {}",
                                        relative,
                                        j + 1,
                                        truncated_text
                                    ));
                                }
                            }

                            // Add the match line
                            let (truncated_text, was_truncated) = truncate_line(line);
                            if was_truncated {
                                *lines_truncated = true;
                            }
                            matches.push(format!("{}:{}: {}", relative, i + 1, truncated_text));

                            // Add context lines after match
                            if context_after > 0 {
                                let end = std::cmp::min(lines.len(), i + context_after + 1);
                                for (j, context_line) in
                                    lines.iter().enumerate().take(end).skip(i + 1)
                                {
                                    let (truncated_text, was_truncated) =
                                        truncate_line(context_line);
                                    if was_truncated {
                                        *lines_truncated = true;
                                    }
                                    matches.push(format!(
                                        "{}-{}- {}",
                                        relative,
                                        j + 1,
                                        truncated_text
                                    ));
                                }
                            }

                            if matches.len() >= max_results {
                                return Ok(());
                            }
                        }
                    }
                }
                Err(_) => {
                    // Skip files we can't read (binary, permissions, etc.)
                }
            }
            return Ok(());
        }

        // Directory: walk entries
        let mut entries = fs::read_dir(current)
            .await
            .map_err(|e| format!("Cannot read directory {}: {}", current.display(), e))?;

        while let Some(entry) = entries
            .next_entry()
            .await
            .map_err(|e| format!("Error reading entry: {}", e))?
        {
            let entry_path = entry.path();

            // Skip hidden files/dirs
            if entry_path
                .file_name()
                .map(|n| n.to_string_lossy().starts_with('.'))
                .unwrap_or(false)
            {
                continue;
            }

            // Skip common non-searchable dirs
            if entry_path.is_dir() {
                let dir_name = entry_path
                    .file_name()
                    .map(|n| n.to_string_lossy().to_string())
                    .unwrap_or_default();
                if matches!(
                    dir_name.as_str(),
                    "node_modules"
                        | "target"
                        | ".git"
                        | "dist"
                        | "build"
                        | "__pycache__"
                        | ".venv"
                        | "venv"
                ) {
                    continue;
                }
            }

            Box::pin(Self::grep_walk(
                root,
                &entry_path,
                re,
                include,
                context_before,
                context_after,
                max_results,
                matches,
                lines_truncated,
            ))
            .await?;
        }

        Ok(())
    }
}

impl Default for GrepTool {
    fn default() -> Self {
        Self::new()
    }
}

#[async_trait]
impl AgentTool for GrepTool {
    fn name(&self) -> &str {
        "grep"
    }

    fn label(&self) -> &str {
        "Grep"
    }

    fn essential(&self) -> bool {
        true
    }
    fn description(&self) -> &str {
        "Search files for a pattern. Returns matching lines with file paths and line numbers. Use literal=true to treat pattern as a literal string. Use context=n to show n lines before and after matches. Long lines are truncated to 500 chars."
    }

    fn parameters_schema(&self) -> Value {
        json!({
            "type": "object",
            "properties": {
                "pattern": {
                    "type": "string",
                    "description": "The pattern to search for (regex by default, or literal string if literal=true)"
                },
                "path": {
                    "type": "string",
                    "description": "The directory or file to search in",
                    "default": "."
                },
                "case_insensitive": {
                    "type": "boolean",
                    "description": "If true, perform case-insensitive search",
                    "default": false
                },
                "literal": {
                    "type": "boolean",
                    "description": "If true, treat pattern as a literal string instead of regex",
                    "default": false
                },
                "context": {
                    "type": "integer",
                    "description": "Number of lines to show before and after each match",
                    "default": 0
                },
                "include": {
                    "type": "string",
                    "description": "Glob pattern to filter files (e.g., '*.rs', '*.ts')"
                },
                "max_results": {
                    "type": "integer",
                    "description": "Maximum number of results to return",
                    "default": 100
                }
            },
            "required": ["pattern"]
        })
    }

    async fn execute(
        &self,
        _tool_call_id: &str,
        params: Value,
        _signal: Option<oneshot::Receiver<()>>,
        ctx: &ToolContext,
    ) -> Result<AgentToolResult, ToolError> {
        let args: GrepArgs =
            serde_json::from_value(params).map_err(|e| format!("invalid params: {e}"))?;
        self.execute_typed(_tool_call_id, args, _signal, ctx).await
    }
}

#[async_trait]
impl TypedTool for GrepTool {
    type Args = GrepArgs;
    async fn execute_typed(
        &self,
        _tool_call_id: &str,
        args: Self::Args,
        _signal: Option<oneshot::Receiver<()>>,
        ctx: &ToolContext,
    ) -> Result<AgentToolResult, ToolError> {
        let path = if args.path.is_empty() {
            "."
        } else {
            &args.path
        };
        let pattern = &args.pattern;
        let case_insensitive = args.case_insensitive;
        let literal = args.literal;
        let context = args.context;
        let include = args.include.as_deref();
        let max_results = args.max_results;

        // ── Internal URL dispatch ──
        if let Some(ref resolver) = ctx.url_resolver
            && resolver.can_resolve(path)
        {
            let resolved = resolver.resolve(path).await?;
            let pattern_str = if literal {
                regex::escape(pattern)
            } else {
                pattern.to_string()
            };
            let re = RegexBuilder::new(&pattern_str)
                .case_insensitive(case_insensitive)
                .build()
                .map_err(|e| format!("Invalid pattern '{}': {}", pattern, e))?;

            let lines: Vec<&str> = resolved.content.lines().collect();
            let mut matches: Vec<String> = Vec::new();
            let mut lines_truncated = false;

            for (i, line) in lines.iter().enumerate() {
                if matches.len() >= max_results {
                    break;
                }
                if re.is_match(line) {
                    if context > 0 && i > 0 {
                        let start = i.saturating_sub(context);
                        for (j, ctx_line) in lines.iter().enumerate().take(i).skip(start) {
                            let (truncated, was_trunc) = truncate_line(ctx_line);
                            if was_trunc {
                                lines_truncated = true;
                            }
                            matches.push(format!("{}-{}- {}", path, j + 1, truncated));
                        }
                    }
                    let (truncated, was_trunc) = truncate_line(line);
                    if was_trunc {
                        lines_truncated = true;
                    }
                    matches.push(format!("{}:{}: {}", path, i + 1, truncated));
                    if context > 0 {
                        let end = std::cmp::min(lines.len(), i + context + 1);
                        for (j, ctx_line) in lines.iter().enumerate().take(end).skip(i + 1) {
                            let (truncated, was_trunc) = truncate_line(ctx_line);
                            if was_trunc {
                                lines_truncated = true;
                            }
                            matches.push(format!("{}-{}- {}", path, j + 1, truncated));
                        }
                    }
                    if matches.len() >= max_results {
                        break;
                    }
                }
            }
            let output = if matches.is_empty() {
                "No matches found".to_string()
            } else {
                format!("Found {} matches:\n{}", matches.len(), matches.join("\n"))
            };
            let mut result = AgentToolResult::success(output);
            if lines_truncated {
                result.metadata = Some(json!({
                    "lines_truncated": true,
                    "message": "Some lines truncated to 500 chars. Use read tool to see full lines."
                }));
            }
            return Ok(result);
        }

        let root = self.root_dir.as_deref().unwrap_or(ctx.root());
        match Self::grep_impl(
            root,
            pattern,
            path,
            case_insensitive,
            literal,
            context,
            context,
            include,
            max_results,
        )
        .await
        {
            Ok((output, lines_truncated)) => {
                let mut result = AgentToolResult::success(output);
                if lines_truncated {
                    result.metadata = Some(json!({
                        "lines_truncated": true,
                        "message": "Some lines truncated to 500 chars. Use read tool to see full lines."
                    }));
                }
                Ok(result)
            }
            Err(e) => Ok(AgentToolResult::error(e)),
        }
    }
}