grep.ts

  1import { spawn, spawnSync } from "node:child_process";
  2import { createInterface } from "node:readline";
  3import { readFileSync, statSync } from "node:fs";
  4import { relative, basename } from "node:path";
  5import { Type } from "@sinclair/typebox";
  6import type { AgentTool } from "@mariozechner/pi-ai";
  7import { resolveToCwd } from "./path-utils.js";
  8import {
  9  DEFAULT_MAX_BYTES,
 10  formatSize,
 11  GREP_MAX_LINE_LENGTH,
 12  truncateHead,
 13  truncateLine,
 14} from "../../util/truncate.js";
 15import { ToolInputError } from "../../util/errors.js";
 16
 17const DEFAULT_LIMIT = 100;
 18
 19const GrepSchema = Type.Object({
 20  pattern: Type.String({ description: "Search pattern (regex or literal string)" }),
 21  path: Type.Optional(Type.String({ description: "Directory or file to search (default: current directory)" })),
 22  glob: Type.Optional(Type.String({ description: "Filter files by glob pattern, e.g. '*.ts'" })),
 23  ignoreCase: Type.Optional(Type.Boolean({ description: "Case-insensitive search (default: false)" })),
 24  literal: Type.Optional(
 25    Type.Boolean({ description: "Treat pattern as literal string instead of regex (default: false)" }),
 26  ),
 27  context: Type.Optional(Type.Number({ description: "Lines of context around each match (default: 0)" })),
 28  limit: Type.Optional(Type.Number({ description: "Maximum matches to return (default: 100)" })),
 29});
 30
 31export const createGrepTool = (workspacePath: string): AgentTool => {
 32  const rgResult = spawnSync("which", ["rg"], { encoding: "utf-8" });
 33  if (rgResult.status !== 0) {
 34    throw new ToolInputError("ripgrep (rg) is not available");
 35  }
 36  const rgPath = rgResult.stdout.trim();
 37
 38  return {
 39    name: "grep",
 40    label: "Grep",
 41    description: `Search file contents for a pattern using ripgrep. Returns up to ${DEFAULT_LIMIT} matches or ${DEFAULT_MAX_BYTES / 1024}KB (whichever is hit first). Lines truncated to ${GREP_MAX_LINE_LENGTH} chars. Respects .gitignore.`,
 42    parameters: GrepSchema as any,
 43    execute: async (_toolCallId: string, params: any) => {
 44      const searchDir: string | undefined = params.path;
 45      const searchPath = resolveToCwd(searchDir || ".", workspacePath);
 46      let isDirectory = false;
 47      try {
 48        isDirectory = statSync(searchPath).isDirectory();
 49      } catch {
 50        throw new ToolInputError(`Path does not exist or is not accessible: ${searchDir || "."}`);
 51      }
 52
 53    const effectiveLimit: number = Math.max(1, Math.floor(params.limit ?? DEFAULT_LIMIT));
 54    const contextLines: number = Math.max(0, Math.floor(params.context ?? 0));
 55
 56    const args: string[] = ["--json", "--line-number", "--color=never", "--hidden"];
 57    if (params.ignoreCase) args.push("--ignore-case");
 58    if (params.literal) args.push("--fixed-strings");
 59    if (params.glob) args.push("--glob", params.glob);
 60    args.push(params.pattern, searchPath);
 61
 62    return new Promise((resolve, reject) => {
 63      const child = spawn(rgPath, args, { stdio: ["ignore", "pipe", "pipe"] });
 64
 65      const rl = createInterface({ input: child.stdout! });
 66      const fileCache = new Map<string, string[]>();
 67
 68      interface MatchEntry {
 69        filePath: string;
 70        lineNumber: number;
 71      }
 72
 73      const matches: MatchEntry[] = [];
 74      let stderrData = "";
 75      let limitReached = false;
 76
 77      child.stderr!.on("data", (chunk: Buffer) => {
 78        stderrData += chunk.toString();
 79      });
 80
 81      rl.on("line", (line: string) => {
 82        if (matches.length >= effectiveLimit) {
 83          if (!limitReached) {
 84            limitReached = true;
 85            child.kill();
 86          }
 87          return;
 88        }
 89
 90        try {
 91          const parsed = JSON.parse(line);
 92          if (parsed.type === "match") {
 93            const filePath: string = parsed.data.path.text;
 94            const lineNumber: number = parsed.data.line_number;
 95            matches.push({ filePath, lineNumber });
 96          }
 97        } catch {
 98          // ignore malformed JSON lines
 99        }
100      });
101
102      child.on("close", (code: number | null) => {
103        if (code !== 0 && code !== 1 && !limitReached) {
104          reject(new Error(`rg exited with code ${code}: ${stderrData.trim()}`));
105          return;
106        }
107
108        if (matches.length === 0) {
109          resolve({
110            content: [{ type: "text", text: "No matches found" }],
111            details: { matches: 0 },
112          });
113          return;
114        }
115
116        const outputLines: string[] = [];
117        let anyLineTruncated = false;
118
119        for (const match of matches) {
120          let lines: string[];
121          if (fileCache.has(match.filePath)) {
122            lines = fileCache.get(match.filePath)!;
123          } else {
124            try {
125              lines = readFileSync(match.filePath, "utf-8").split(/\r?\n/);
126              fileCache.set(match.filePath, lines);
127            } catch {
128              lines = [];
129            }
130          }
131
132          const relPath = isDirectory
133            ? relative(searchPath, match.filePath)
134            : basename(match.filePath);
135
136          const startLine = Math.max(0, match.lineNumber - 1 - contextLines);
137          const endLine = Math.min(lines.length, match.lineNumber + contextLines);
138
139          for (let i = startLine; i < endLine; i++) {
140            const lineContent = lines[i] ?? "";
141            const { text: truncatedContent, wasTruncated } = truncateLine(lineContent);
142            if (wasTruncated) anyLineTruncated = true;
143
144            const lineNum = i + 1;
145            if (lineNum === match.lineNumber) {
146              outputLines.push(`${relPath}:${lineNum}: ${truncatedContent}`);
147            } else {
148              outputLines.push(`${relPath}-${lineNum}- ${truncatedContent}`);
149            }
150          }
151        }
152
153        const rawOutput = outputLines.join("\n");
154        const truncation = truncateHead(rawOutput, { maxLines: Number.MAX_SAFE_INTEGER });
155
156        const notices: string[] = [];
157        if (limitReached) {
158          notices.push(
159            `${effectiveLimit} matches limit reached. Use limit=${effectiveLimit * 2} for more, or refine pattern`,
160          );
161        }
162        if (truncation.truncated) {
163          notices.push(`${formatSize(DEFAULT_MAX_BYTES)} limit reached`);
164        }
165        if (anyLineTruncated) {
166          notices.push(
167            `Some lines truncated to ${GREP_MAX_LINE_LENGTH} chars. Use read tool to see full lines`,
168          );
169        }
170
171        let output = truncation.content;
172        if (notices.length > 0) {
173          output += `\n\n[${notices.join(". ")}]`;
174        }
175
176        resolve({
177          content: [{ type: "text", text: output }],
178          details: {
179            matches: matches.length,
180            ...(limitReached ? { matchLimitReached: effectiveLimit } : {}),
181            ...(truncation.truncated ? { truncation } : {}),
182            ...(anyLineTruncated ? { linesTruncated: true } : {}),
183          },
184        });
185      });
186    });
187    },
188  };
189};