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/edit_file.py

@@ -0,0 +1,65 @@
+"""edit_file.py – tool for str_replace editing of files 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_edit_file_tool(sandbox: "DockerSandbox") -> BaseTool:
+    """Return an edit_file tool bound to *sandbox*."""
+
+    @tool
+    def edit_file(path: str, old_str: str, new_str: str) -> str:
+        """
+        Replace the first exact occurrence of *old_str* with *new_str* in *path*.
+
+        This is the standard ``str_replace`` editing primitive: read the file,
+        find the unique snippet you want to change, and supply the replacement.
+
+        Rules:
+        - *old_str* must match **exactly** (including whitespace and indentation).
+        - *old_str* must appear **at least once**; the tool returns an error if it
+          is not found.
+        - If *old_str* appears more than once the tool refuses and asks you to
+          provide more surrounding context to make it unique.
+        - To insert text without removing anything, set *old_str* to a line that
+          will remain and include it verbatim in *new_str* (i.e. keep the anchor
+          line and add your new lines around it).
+        - To delete a block, set *new_str* to an empty string ``""``.
+
+        Returns a confirmation with the number of lines affected, or an error.
+        """
+        logger.debug("Editing file inside sandbox: {!r}", path)
+        exit_code, content = sandbox.exec(f"cat -- {quote(path)}")
+        if exit_code != 0:
+            return f"[ERROR reading {path!r} for edit] {content.strip()}"
+
+        count = content.count(old_str)
+        if count == 0:
+            return (
+                f"[ERROR] old_str not found in {path!r}. "
+                "Check that whitespace and indentation match exactly."
+            )
+        if count > 1:
+            return (
+                f"[ERROR] old_str appears {count} times in {path!r}. "
+                "Provide more surrounding context to make it unique."
+            )
+
+        new_content = content.replace(old_str, new_str, 1)
+        old_lines = old_str.count("\n") + 1
+        new_lines = new_str.count("\n") + 1 if new_str else 0
+        try:
+            sandbox.write_file(path, new_content)
+        except Exception as exc:
+            return f"[ERROR writing {path!r} after edit] {exc}"
+        return f"[OK] Replaced {old_lines} line(s) with {new_lines} line(s) in {path}"
+
+    return edit_file