feat: Initial library extraction from PIN LLM benchmark

DockerSandbox + LangChain file/shell tools extracted into a standalone package.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

docker_agent_sandbox/tools/grep.py

@@ -0,0 +1,45 @@
+"""grep.py – tool for searching file contents inside the sandbox."""
+
+from __future__ import annotations
+
+from shlex import quote
+from typing import TYPE_CHECKING
+
+from langchain_core.tools import BaseTool, tool
+from loguru import logger
+
+if TYPE_CHECKING:
+    from docker_agent_sandbox.sandbox import DockerSandbox
+
+
+def make_grep_tool(sandbox: "DockerSandbox") -> BaseTool:
+    """Return a grep tool bound to *sandbox*."""
+
+    @tool
+    def grep(pattern: str, path: str, recursive: bool = False) -> str:
+        """
+        Search for *pattern* (extended regex) in *path*.
+
+        *path* can be a file or a directory; when *path* is a directory,
+        *recursive* must be ``True``.  Returns matching lines with file names
+        and line numbers, or an error message.
+
+        Useful for locating strings, symbol names, or byte sequences in
+        binaries and text files.
+        """
+        logger.debug(
+            "Grepping inside sandbox: pattern={!r} path={!r} recursive={}",
+            pattern,
+            path,
+            recursive,
+        )
+        flags = "-rn" if recursive else "-n"
+        exit_code, output = sandbox.exec(
+            f"grep -E {flags} -- {quote(pattern)} {quote(path)} 2>&1"
+        )
+        # grep exits 1 when no matches — that is not an error
+        if exit_code not in (0, 1):
+            return f"[ERROR grepping {path!r}] {output.strip()}"
+        return output.strip() or "[no matches found]"
+
+    return grep