lm-deluge 0.0.67__py3-none-any.whl → 0.0.90__py3-none-any.whl

lm_deluge/tool/prefab/filesystem.py

@@ -0,0 +1,1711 @@
+from __future__ import annotations
+
+import io
+import json
+import os
+import random
+import re
+import time
+import zipfile
+from collections.abc import Callable, Iterable, Sequence
+from dataclasses import dataclass
+from functools import partial
+from pathlib import Path
+from typing import Any, Dict, Literal, Optional, Protocol
+
+from pydantic import BaseModel, Field
+
+from .. import Tool
+
+FS_DESCRIPTION = """Interact with an isolated virtual filesystem that belongs to this session.
+
+Paths are always relative to the workspace root and use forward slashes. Use this tool to:
+- inspect files with optional line ranges
+- create, overwrite, or append to files
+- delete files or folders from the workspace
+- list directory contents
+- search for text across the workspace using regular expressions
+- apply OpenAI-style apply_patch operations (create/update/delete)
+
+This virtual filesystem is safe-by-construction. Paths that try to escape the workspace
+or reference missing files will raise clear errors."""
+
+
+class WorkspaceBackend(Protocol):
+    """Abstract filesystem operations used by FilesystemManager."""
+
+    def read_file(self, path: str) -> str: ...
+
+    def write_file(self, path: str, content: str, *, overwrite: bool) -> None: ...
+
+    def append_file(self, path: str, content: str) -> None: ...
+
+    def delete_path(self, path: str) -> None: ...
+
+    def list_dir(self, path: str, recursive: bool) -> list[dict[str, Any]]: ...
+
+    def grep(
+        self, pattern: str, path: str | None, limit: int
+    ) -> list[dict[str, Any]]: ...
+
+
+def _normalize_path(path: str | None, *, allow_root: bool = False) -> str:
+    """Normalize user-provided paths and prevent directory traversal."""
+    if path is None or path.strip() == "":
+        if allow_root:
+            return "."
+        raise ValueError("Path is required")
+
+    raw = path.strip()
+    if raw.startswith("/"):
+        raw = raw.lstrip("/")
+
+    parts: list[str] = []
+    for part in raw.split("/"):
+        if part in ("", "."):
+            continue
+        if part == "..":
+            if parts:
+                parts.pop()
+            else:
+                raise ValueError("Path traversal outside the workspace is not allowed")
+            continue
+        parts.append(part)
+
+    normalized = "/".join(parts)
+    if normalized:
+        return normalized
+    if allow_root:
+        return "."
+    raise ValueError("Path must reference a file inside the workspace")
+
+
+class InMemoryWorkspaceBackend:
+    """Simple backend that stores files in memory."""
+
+    def __init__(self, files: dict[str, str] | None = None):
+        self._files: Dict[str, str] = {}
+        if files:
+            for path, content in files.items():
+                key = _normalize_path(path)
+                self._files[key] = content
+
+    def read_file(self, path: str) -> str:
+        key = _normalize_path(path)
+        if key not in self._files:
+            raise FileNotFoundError(f"{key} does not exist")
+        return self._files[key]
+
+    def write_file(self, path: str, content: str, *, overwrite: bool) -> None:
+        key = _normalize_path(path)
+        if not overwrite and key in self._files:
+            raise FileExistsError(f"{key} already exists")
+        self._files[key] = content
+
+    def append_file(self, path: str, content: str) -> None:
+        key = _normalize_path(path)
+        if key in self._files:
+            self._files[key] = f"{self._files[key]}{content}"
+        else:
+            self._files[key] = content
+
+    def delete_path(self, path: str) -> None:
+        key = _normalize_path(path, allow_root=True)
+        if key == ".":
+            self._files.clear()
+            return
+        if key in self._files:
+            del self._files[key]
+            return
+
+        prefix = f"{key}/"
+        targets = [p for p in self._files if p.startswith(prefix)]
+        if not targets:
+            raise FileNotFoundError(f"{key} does not exist")
+        for target in targets:
+            del self._files[target]
+
+    def list_dir(self, path: str, recursive: bool) -> list[dict[str, Any]]:
+        key = _normalize_path(path, allow_root=True)
+        if key != "." and key in self._files and not recursive:
+            # Listing a file path shows metadata for that file.
+            return [self._format_file_entry(key)]
+
+        prefix = "" if key == "." else f"{key}/"
+        entries: list[dict[str, Any]] = []
+
+        if key != "." and not any(
+            p == key or p.startswith(prefix) for p in self._files
+        ):
+            raise FileNotFoundError(f"{key} does not exist")
+
+        if recursive:
+            for file_path in sorted(self._files):
+                if not (file_path == key or file_path.startswith(prefix)):
+                    continue
+                entries.append(self._format_file_entry(file_path))
+            return entries
+
+        seen_dirs: set[str] = set()
+        for file_path in sorted(self._files):
+            if not (file_path == key or file_path.startswith(prefix)):
+                continue
+            remainder = file_path[len(prefix) :]
+            if remainder == "":
+                entries.append(self._format_file_entry(file_path))
+                continue
+            head, _, tail = remainder.partition("/")
+            if tail:
+                dir_path = head if key == "." else f"{key}/{head}"
+                if dir_path not in seen_dirs:
+                    entries.append(
+                        {"path": dir_path, "type": "directory", "size": None}
+                    )
+                    seen_dirs.add(dir_path)
+            else:
+                entries.append(self._format_file_entry(file_path))
+        return entries
+
+    def grep(self, pattern: str, path: str | None, limit: int) -> list[dict[str, Any]]:
+        regex = re.compile(pattern)
+        key = _normalize_path(path, allow_root=True) if path is not None else "."
+        prefix = "" if key == "." else f"{key}/"
+        matches: list[dict[str, Any]] = []
+
+        for file_path in sorted(self._files):
+            if not (file_path == key or file_path.startswith(prefix)):
+                continue
+            content = self._files[file_path]
+            for line_no, line in enumerate(content.splitlines(), start=1):
+                if regex.search(line):
+                    matches.append(
+                        {"path": file_path, "line": line_no, "text": line.strip()}
+                    )
+                    if len(matches) >= limit:
+                        return matches
+        if (
+            key != "."
+            and key not in self._files
+            and not any(p.startswith(prefix) for p in self._files)
+        ):
+            raise FileNotFoundError(f"{key} does not exist")
+        return matches
+
+    def _format_file_entry(self, path: str) -> dict[str, Any]:
+        content = self._files[path]
+        if content == "":
+            line_count = 0
+        else:
+            line_count = content.count("\n") + (0 if content.endswith("\n") else 1)
+        return {
+            "path": path,
+            "type": "file",
+            "size": len(content),
+            "line_count": max(line_count, 0),
+        }
+
+
+FsCommand = Literal[
+    "read_file",
+    "write_file",
+    "delete_path",
+    "list_dir",
+    "grep",
+    "apply_patch",
+]
+ALL_COMMANDS: tuple[FsCommand, ...] = (
+    "read_file",
+    "write_file",
+    "delete_path",
+    "list_dir",
+    "grep",
+    "apply_patch",
+)
+
+
+class ApplyPatchOperation(BaseModel):
+    """Subset of OpenAI apply_patch operation payload."""
+
+    type: Literal["create_file", "update_file", "delete_file"] = Field(
+        description="Type of patch operation to perform."
+    )
+    path: str = Field(description="Path to the file being modified.")
+    diff: str | None = Field(
+        default=None,
+        description="V4A diff to apply for create/update operations.",
+    )
+
+    @property
+    def requires_diff(self) -> bool:
+        return self.type in {"create_file", "update_file"}
+
+
+class FilesystemParams(BaseModel):
+    """Schema describing filesystem tool calls."""
+
+    command: FsCommand = Field(
+        description="Filesystem operation to perform (read_file, write_file, delete_path, list_dir, grep)"
+    )
+    path: Optional[str] = Field(
+        default=None,
+        description="Path to operate on, relative to workspace root. Use '.' for the root directory.",
+    )
+    start_line: Optional[int] = Field(
+        default=None,
+        description="1-based inclusive start line when reading a file. Leave unset to read from the beginning.",
+        ge=1,
+    )
+    end_line: Optional[int] = Field(
+        default=None,
+        description="1-based inclusive end line when reading a file. Leave unset to read through the end.",
+        ge=1,
+    )
+    content: Optional[str] = Field(
+        default=None,
+        description="Content to write when using write_file.",
+    )
+    mode: Optional[Literal["overwrite", "append", "create_if_missing"]] = Field(
+        default="overwrite",
+        description="How to write content. Overwrite replaces the file, append adds to the end, create_if_missing leaves existing files untouched.",
+    )
+    recursive: Optional[bool] = Field(
+        default=None,
+        description="When listing directories, set to true to recurse.",
+    )
+    pattern: Optional[str] = Field(
+        default=None,
+        description="Regular expression pattern to search for when using grep.",
+    )
+    max_results: Optional[int] = Field(
+        default=50,
+        description="Maximum number of grep matches to return.",
+        ge=1,
+    )
+    operation: ApplyPatchOperation | None = Field(
+        default=None,
+        description=(
+            "When using command='apply_patch', include an operation matching the "
+            "OpenAI apply_patch_call payload (type, path, diff)."
+        ),
+    )
+
+
+class FilesystemManager:
+    """Expose a TodoManager-style tool for interacting with a workspace."""
+
+    def __init__(
+        self,
+        backend: WorkspaceBackend | None = None,
+        *,
+        tool_name: str = "filesystem",
+    ):
+        self.backend = backend or InMemoryWorkspaceBackend()
+        self.tool_name = tool_name
+        self._tool_cache: dict[tuple[str, ...], list[Tool]] = {}
+
+    def _handle_read(
+        self, path: str, start_line: Optional[int], end_line: Optional[int]
+    ) -> dict[str, Any]:
+        content = self.backend.read_file(path)
+        total_lines = len(content.splitlines()) or (0 if content == "" else 1)
+        start = start_line or 1
+        end = end_line or total_lines
+        if end < start:
+            if not (total_lines == 0 and end_line is None and start == 1):
+                raise ValueError("end_line must be greater than or equal to start_line")
+
+        if start == 1 and end >= total_lines:
+            snippet = content
+        else:
+            lines = content.splitlines()
+            snippet = "\n".join(lines[start - 1 : end])
+
+        return {
+            "path": path,
+            "start_line": start,
+            "end_line": end,
+            "content": snippet,
+            "total_lines": total_lines,
+            "character_count": len(content),
+        }
+
+    def _handle_write(
+        self, path: str, content: str, mode: Optional[str]
+    ) -> dict[str, Any]:
+        write_mode = mode or "overwrite"
+        if write_mode == "overwrite":
+            self.backend.write_file(path, content, overwrite=True)
+        elif write_mode == "append":
+            self.backend.append_file(path, content)
+        elif write_mode == "create_if_missing":
+            try:
+                self.backend.write_file(path, content, overwrite=False)
+            except FileExistsError:
+                pass
+        else:
+            raise ValueError(f"Unsupported write mode: {write_mode}")
+        return {"path": path, "status": "ok", "mode": write_mode}
+
+    def _handle_delete(self, path: str) -> dict[str, Any]:
+        self.backend.delete_path(path)
+        return {"path": path, "status": "ok"}
+
+    def _handle_list(
+        self, path: Optional[str], recursive: Optional[bool]
+    ) -> dict[str, Any]:
+        listing = self.backend.list_dir(path or ".", recursive=bool(recursive))
+        return {"path": path or ".", "recursive": bool(recursive), "entries": listing}
+
+    def _handle_grep(
+        self, pattern: str, path: Optional[str], limit: Optional[int]
+    ) -> dict[str, Any]:
+        max_results = limit or 50
+        matches = self.backend.grep(pattern, path=path, limit=max_results)
+        return {
+            "pattern": pattern,
+            "path": path,
+            "max_results": max_results,
+            "matches": matches,
+        }
+
+    def _handle_apply_patch(self, operation: ApplyPatchOperation) -> dict[str, Any]:
+        if operation.requires_diff and not operation.diff:
+            raise ValueError("diff is required for create_file and update_file")
+
+        if operation.type == "delete_file":
+            self.backend.delete_path(operation.path)
+            return {"path": operation.path, "operation": "delete_file"}
+
+        assert operation.diff is not None  # for type checkers
+        if operation.type == "create_file":
+            new_content = apply_diff("", operation.diff, mode="create")
+            self.backend.write_file(operation.path, new_content, overwrite=False)
+            return {"path": operation.path, "operation": "create_file"}
+
+        if operation.type == "update_file":
+            current = self.backend.read_file(operation.path)
+            new_content = apply_diff(current, operation.diff, mode="default")
+            self.backend.write_file(operation.path, new_content, overwrite=True)
+            return {"path": operation.path, "operation": "update_file"}
+
+        raise ValueError(f"Unsupported patch operation: {operation.type}")
+
+    def _filesystem_tool(self, allowed_commands: set[FsCommand], **kwargs: Any) -> str:
+        params = FilesystemParams.model_validate(kwargs)
+
+        try:
+            if params.command not in allowed_commands:
+                raise ValueError(
+                    f"The '{params.command}' command is disabled for this tool instance"
+                )
+            if params.command == "read_file":
+                if not params.path:
+                    raise ValueError("path is required for read_file")
+                result = self._handle_read(
+                    params.path, params.start_line, params.end_line
+                )
+            elif params.command == "write_file":
+                if params.path is None or params.content is None:
+                    raise ValueError("path and content are required for write_file")
+                result = self._handle_write(params.path, params.content, params.mode)
+            elif params.command == "delete_path":
+                if not params.path:
+                    raise ValueError("path is required for delete_path")
+                result = self._handle_delete(params.path)
+            elif params.command == "list_dir":
+                result = self._handle_list(params.path, params.recursive)
+            elif params.command == "grep":
+                if not params.pattern:
+                    raise ValueError("pattern is required for grep")
+                result = self._handle_grep(
+                    params.pattern, params.path, params.max_results
+                )
+            elif params.command == "apply_patch":
+                if params.operation is None:
+                    raise ValueError("operation is required for apply_patch")
+                result = self._handle_apply_patch(params.operation)
+            else:
+                raise ValueError(f"Unknown command: {params.command}")
+            return json.dumps({"ok": True, "result": result}, indent=2)
+        except Exception as exc:
+            return json.dumps(
+                {"ok": False, "error": type(exc).__name__, "message": str(exc)},
+                indent=2,
+            )
+
+    def get_tools(self, *, exclude: Iterable[FsCommand] | None = None) -> list[Tool]:
+        exclude_set = set(exclude or [])
+        unknown = exclude_set.difference(ALL_COMMANDS)
+        if unknown:
+            raise ValueError(f"Unknown commands in exclude list: {sorted(unknown)}")
+
+        allowed = tuple(cmd for cmd in ALL_COMMANDS if cmd not in exclude_set)
+        if not allowed:
+            raise ValueError("Cannot exclude every filesystem command")
+
+        cache_key = allowed
+        if cache_key in self._tool_cache:
+            return self._tool_cache[cache_key]
+
+        allowed_set = set(allowed)
+        schema = FilesystemParams.model_json_schema(ref_template="#/$defs/{model}")
+        if (
+            "properties" in schema
+            and "command" in schema["properties"]
+            and isinstance(schema["properties"]["command"], dict)
+        ):
+            schema["properties"]["command"]["enum"] = list(allowed)
+
+        tool = Tool(
+            name=self.tool_name,
+            description=FS_DESCRIPTION,
+            parameters=schema.get("properties", {}),
+            required=schema.get("required", []),
+            definitions=schema.get("$defs"),
+            run=partial(self._filesystem_tool, allowed_set),  # type: ignore
+        )
+
+        self._tool_cache[cache_key] = [tool]
+        return [tool]
+
+    def dump(
+        self,
+        destination: str | os.PathLike[str],
+        *,
+        as_zip: bool = False,
+    ) -> list[str]:
+        """
+        Copy the virtual workspace to the given filesystem path.
+
+        Args:
+            destination: Path to write to. If as_zip=True, this should be a .zip file path.
+            as_zip: If True, write as a zip archive instead of a directory.
+
+        Returns:
+            List of file paths that were written.
+        """
+        entries = self.backend.list_dir(".", recursive=True)
+        written: list[str] = []
+
+        if as_zip:
+            target_path = Path(destination)
+            target_path.parent.mkdir(parents=True, exist_ok=True)
+
+            with zipfile.ZipFile(target_path, "w", zipfile.ZIP_DEFLATED) as zf:
+                for entry in entries:
+                    if entry.get("type") != "file":
+                        continue
+                    rel_path = entry["path"]
+                    content = self.backend.read_file(rel_path)
+                    zf.writestr(rel_path, content)
+                    written.append(rel_path)
+        else:
+            target_root = Path(destination)
+            if target_root.exists() and not target_root.is_dir():
+                raise NotADirectoryError(f"{target_root} exists and is not a directory")
+            target_root.mkdir(parents=True, exist_ok=True)
+
+            for entry in entries:
+                if entry.get("type") != "file":
+                    continue
+                rel_path = entry["path"]
+                destination_path = target_root.joinpath(*rel_path.split("/"))
+                destination_path.parent.mkdir(parents=True, exist_ok=True)
+                destination_path.write_text(self.backend.read_file(rel_path))
+                written.append(rel_path)
+
+        return sorted(written)
+
+    @classmethod
+    def from_dir(
+        cls,
+        source: str | os.PathLike[str],
+        *,
+        max_files: int = 100,
+        tool_name: str = "filesystem",
+    ) -> "FilesystemManager":
+        """
+        Create a FilesystemManager pre-populated with files from a directory.
+
+        Args:
+            source: Path to the directory to load files from.
+            max_files: Maximum number of files to load (default 100).
+            tool_name: Name for the filesystem tool.
+
+        Returns:
+            A new FilesystemManager with the files loaded into memory.
+
+        Raises:
+            ValueError: If more than max_files files are found.
+            NotADirectoryError: If source is not a directory.
+        """
+        source_path = Path(source)
+        if not source_path.is_dir():
+            raise NotADirectoryError(f"{source_path} is not a directory")
+
+        files: dict[str, str] = {}
+        file_count = 0
+
+        for file_path in source_path.rglob("*"):
+            if not file_path.is_file():
+                continue
+
+            file_count += 1
+            if file_count > max_files:
+                raise ValueError(
+                    f"Directory contains more than {max_files} files. "
+                    f"Increase max_files or use a smaller directory."
+                )
+
+            rel_path = file_path.relative_to(source_path).as_posix()
+
+            # Try to read as text, skip binary files
+            try:
+                content = file_path.read_text()
+                files[rel_path] = content
+            except UnicodeDecodeError:
+                # Skip binary files
+                continue
+
+        backend = InMemoryWorkspaceBackend(files)
+        return cls(backend=backend, tool_name=tool_name)
+
+    @classmethod
+    def from_zip(
+        cls,
+        source: str | os.PathLike[str] | io.BytesIO,
+        *,
+        max_files: int = 100,
+        tool_name: str = "filesystem",
+    ) -> "FilesystemManager":
+        """
+        Create a FilesystemManager pre-populated with files from a zip archive.
+
+        Args:
+            source: Path to the zip file, or a BytesIO containing zip data.
+            max_files: Maximum number of files to load (default 100).
+            tool_name: Name for the filesystem tool.
+
+        Returns:
+            A new FilesystemManager with the files loaded into memory.
+
+        Raises:
+            ValueError: If more than max_files files are found.
+            zipfile.BadZipFile: If the source is not a valid zip file.
+        """
+        files: dict[str, str] = {}
+        file_count = 0
+
+        with zipfile.ZipFile(source, "r") as zf:
+            for info in zf.infolist():
+                # Skip directories
+                if info.is_dir():
+                    continue
+
+                file_count += 1
+                if file_count > max_files:
+                    raise ValueError(
+                        f"Zip archive contains more than {max_files} files. "
+                        f"Increase max_files or use a smaller archive."
+                    )
+
+                # Normalize path (remove leading slashes, handle Windows paths)
+                rel_path = info.filename.lstrip("/").replace("\\", "/")
+                if not rel_path:
+                    continue
+
+                # Try to read as text, skip binary files
+                try:
+                    content = zf.read(info.filename).decode("utf-8")
+                    files[rel_path] = content
+                except UnicodeDecodeError:
+                    # Skip binary files
+                    continue
+
+        backend = InMemoryWorkspaceBackend(files)
+        return cls(backend=backend, tool_name=tool_name)
+
+
+ApplyDiffMode = Literal["default", "create"]
+
+
+@dataclass
+class Chunk:
+    orig_index: int
+    del_lines: list[str]
+    ins_lines: list[str]
+
+
+@dataclass
+class ParserState:
+    lines: list[str]
+    index: int = 0
+    fuzz: int = 0
+
+
+@dataclass
+class ParsedUpdateDiff:
+    chunks: list[Chunk]
+    fuzz: int
+
+
+@dataclass
+class ReadSectionResult:
+    next_context: list[str]
+    section_chunks: list[Chunk]
+    end_index: int
+    eof: bool
+
+
+END_PATCH = "*** End Patch"
+END_FILE = "*** End of File"
+SECTION_TERMINATORS = [
+    END_PATCH,
+    "*** Update File:",
+    "*** Delete File:",
+    "*** Add File:",
+]
+END_SECTION_MARKERS = [*SECTION_TERMINATORS, END_FILE]
+
+
+def apply_diff(input_text: str, diff: str, mode: ApplyDiffMode = "default") -> str:
+    """Apply a V4A diff to the provided text."""
+
+    diff_lines = _normalize_diff_lines(diff)
+    if mode == "create":
+        return _parse_create_diff(diff_lines)
+
+    parsed = _parse_update_diff(diff_lines, input_text)
+    return _apply_chunks(input_text, parsed.chunks)
+
+
+def _normalize_diff_lines(diff: str) -> list[str]:
+    lines = [line.rstrip("\r") for line in re.split(r"\r?\n", diff)]
+    if lines and lines[-1] == "":
+        lines.pop()
+    return lines
+
+
+def _is_done(state: ParserState, prefixes: Sequence[str]) -> bool:
+    if state.index >= len(state.lines):
+        return True
+    if any(state.lines[state.index].startswith(prefix) for prefix in prefixes):
+        return True
+    return False
+
+
+def _read_str(state: ParserState, prefix: str) -> str:
+    if state.index >= len(state.lines):
+        return ""
+    current = state.lines[state.index]
+    if current.startswith(prefix):
+        state.index += 1
+        return current[len(prefix) :]
+    return ""
+
+
+def _parse_create_diff(lines: list[str]) -> str:
+    parser = ParserState(lines=[*lines, END_PATCH])
+    output: list[str] = []
+
+    while not _is_done(parser, SECTION_TERMINATORS):
+        if parser.index >= len(parser.lines):
+            break
+        line = parser.lines[parser.index]
+        parser.index += 1
+        if not line.startswith("+"):
+            raise ValueError(f"Invalid Add File Line: {line}")
+        output.append(line[1:])
+
+    return "\n".join(output)
+
+
+def _parse_update_diff(lines: list[str], input_text: str) -> ParsedUpdateDiff:
+    parser = ParserState(lines=[*lines, END_PATCH])
+    input_lines = input_text.split("\n")
+    chunks: list[Chunk] = []
+    cursor = 0
+
+    while not _is_done(parser, END_SECTION_MARKERS):
+        anchor = _read_str(parser, "@@ ")
+        has_bare_anchor = (
+            anchor == ""
+            and parser.index < len(parser.lines)
+            and parser.lines[parser.index] == "@@"
+        )
+        if has_bare_anchor:
+            parser.index += 1
+
+        if not (anchor or has_bare_anchor or cursor == 0):
+            current_line = (
+                parser.lines[parser.index] if parser.index < len(parser.lines) else ""
+            )
+            raise ValueError(f"Invalid Line:\n{current_line}")
+
+        if anchor.strip():
+            cursor = _advance_cursor_to_anchor(anchor, input_lines, cursor, parser)
+
+        section = _read_section(parser.lines, parser.index)
+        find_result = _find_context(
+            input_lines, section.next_context, cursor, section.eof
+        )
+        if find_result.new_index == -1:
+            ctx_text = "\n".join(section.next_context)
+            if section.eof:
+                raise ValueError(f"Invalid EOF Context {cursor}:\n{ctx_text}")
+            raise ValueError(f"Invalid Context {cursor}:\n{ctx_text}")
+
+        cursor = find_result.new_index + len(section.next_context)
+        parser.fuzz += find_result.fuzz
+        parser.index = section.end_index
+
+        for ch in section.section_chunks:
+            chunks.append(
+                Chunk(
+                    orig_index=ch.orig_index + find_result.new_index,
+                    del_lines=list(ch.del_lines),
+                    ins_lines=list(ch.ins_lines),
+                )
+            )
+
+    return ParsedUpdateDiff(chunks=chunks, fuzz=parser.fuzz)
+
+
+def _advance_cursor_to_anchor(
+    anchor: str,
+    input_lines: list[str],
+    cursor: int,
+    parser: ParserState,
+) -> int:
+    found = False
+
+    if not any(line == anchor for line in input_lines[:cursor]):
+        for i in range(cursor, len(input_lines)):
+            if input_lines[i] == anchor:
+                cursor = i + 1
+                found = True
+                break
+
+    if not found and not any(
+        line.strip() == anchor.strip() for line in input_lines[:cursor]
+    ):
+        for i in range(cursor, len(input_lines)):
+            if input_lines[i].strip() == anchor.strip():
+                cursor = i + 1
+                parser.fuzz += 1
+                found = True
+                break
+
+    return cursor
+
+
+def _read_section(lines: list[str], start_index: int) -> ReadSectionResult:
+    context: list[str] = []
+    del_lines: list[str] = []
+    ins_lines: list[str] = []
+    section_chunks: list[Chunk] = []
+    mode: Literal["keep", "add", "delete"] = "keep"
+    index = start_index
+    orig_index = index
+
+    while index < len(lines):
+        raw = lines[index]
+        if (
+            raw.startswith("@@")
+            or raw.startswith(END_PATCH)
+            or raw.startswith("*** Update File:")
+            or raw.startswith("*** Delete File:")
+            or raw.startswith("*** Add File:")
+            or raw.startswith(END_FILE)
+        ):
+            break
+        if raw == "***":
+            break
+        if raw.startswith("***"):
+            raise ValueError(f"Invalid Line: {raw}")
+
+        index += 1
+        last_mode = mode
+        line = raw if raw else " "
+        prefix = line[0]
+        if prefix == "+":
+            mode = "add"
+        elif prefix == "-":
+            mode = "delete"
+        elif prefix == " ":
+            mode = "keep"
+        else:
+            raise ValueError(f"Invalid Line: {line}")
+
+        line_content = line[1:]
+        switching_to_context = mode == "keep" and last_mode != mode
+        if switching_to_context and (del_lines or ins_lines):
+            section_chunks.append(
+                Chunk(
+                    orig_index=len(context) - len(del_lines),
+                    del_lines=list(del_lines),
+                    ins_lines=list(ins_lines),
+                )
+            )
+            del_lines = []
+            ins_lines = []
+
+        if mode == "delete":
+            del_lines.append(line_content)
+            context.append(line_content)
+        elif mode == "add":
+            ins_lines.append(line_content)
+        else:
+            context.append(line_content)
+
+    if del_lines or ins_lines:
+        section_chunks.append(
+            Chunk(
+                orig_index=len(context) - len(del_lines),
+                del_lines=list(del_lines),
+                ins_lines=list(ins_lines),
+            )
+        )
+
+    if index < len(lines) and lines[index] == END_FILE:
+        return ReadSectionResult(context, section_chunks, index + 1, True)
+
+    if index == orig_index:
+        next_line = lines[index] if index < len(lines) else ""
+        raise ValueError(f"Nothing in this section - index={index} {next_line}")
+
+    return ReadSectionResult(context, section_chunks, index, False)
+
+
+@dataclass
+class ContextMatch:
+    new_index: int
+    fuzz: int
+
+
+def _find_context(
+    lines: list[str], context: list[str], start: int, eof: bool
+) -> ContextMatch:
+    if eof:
+        end_start = max(0, len(lines) - len(context))
+        end_match = _find_context_core(lines, context, end_start)
+        if end_match.new_index != -1:
+            return end_match
+        fallback = _find_context_core(lines, context, start)
+        return ContextMatch(new_index=fallback.new_index, fuzz=fallback.fuzz + 10000)
+    return _find_context_core(lines, context, start)
+
+
+def _find_context_core(
+    lines: list[str], context: list[str], start: int
+) -> ContextMatch:
+    if not context:
+        return ContextMatch(new_index=start, fuzz=0)
+
+    for i in range(start, len(lines)):
+        if _equals_slice(lines, context, i, lambda value: value):
+            return ContextMatch(new_index=i, fuzz=0)
+    for i in range(start, len(lines)):
+        if _equals_slice(lines, context, i, lambda value: value.rstrip()):
+            return ContextMatch(new_index=i, fuzz=1)
+    for i in range(start, len(lines)):
+        if _equals_slice(lines, context, i, lambda value: value.strip()):
+            return ContextMatch(new_index=i, fuzz=100)
+
+    return ContextMatch(new_index=-1, fuzz=0)
+
+
+def _equals_slice(
+    source: list[str], target: list[str], start: int, map_fn: Callable[[str], str]
+) -> bool:
+    if start + len(target) > len(source):
+        return False
+    for offset, target_value in enumerate(target):
+        if map_fn(source[start + offset]) != map_fn(target_value):
+            return False
+    return True
+
+
+def _apply_chunks(input_text: str, chunks: list[Chunk]) -> str:
+    orig_lines = input_text.split("\n")
+    dest_lines: list[str] = []
+    cursor = 0
+
+    for chunk in chunks:
+        if chunk.orig_index > len(orig_lines):
+            raise ValueError(
+                f"apply_diff: chunk.origIndex {chunk.orig_index} > input length {len(orig_lines)}"
+            )
+        if cursor > chunk.orig_index:
+            raise ValueError(
+                f"apply_diff: overlapping chunk at {chunk.orig_index} (cursor {cursor})"
+            )
+
+        dest_lines.extend(orig_lines[cursor : chunk.orig_index])
+        cursor = chunk.orig_index
+
+        if chunk.ins_lines:
+            dest_lines.extend(chunk.ins_lines)
+
+        cursor += len(chunk.del_lines)
+
+    dest_lines.extend(orig_lines[cursor:])
+    return "\n".join(dest_lines)
+
+
+# S3 filesystem description
+S3_FS_DESCRIPTION = """Interact with a remote S3-backed filesystem that supports safe concurrent access.
+
+This filesystem is backed by Amazon S3 with optimistic concurrency control, meaning multiple
+agents can safely read and write to the same workspace without conflicts. If a write conflict
+occurs (another agent modified the file), the operation will automatically retry.
+
+Paths are always relative to the workspace root and use forward slashes. Use this tool to:
+- inspect files with optional line ranges
+- create, overwrite, or append to files (with automatic conflict resolution)
+- delete files or folders from the workspace
+- list directory contents
+- search for text across the workspace using regular expressions
+
+This filesystem is safe for distributed use - conflicts are automatically detected and resolved."""
+
+
+@dataclass
+class S3FileMetadata:
+    """Metadata for a file stored in S3."""
+
+    key: str
+    etag: str
+    size: int
+    last_modified: float
+
+
+@dataclass
+class ConflictError(Exception):
+    """Raised when a write conflict occurs due to concurrent modification."""
+
+    key: str
+    expected_etag: str | None
+    actual_etag: str | None
+    message: str = ""
+
+    def __str__(self) -> str:
+        if self.message:
+            return self.message
+        return f"Conflict writing {self.key}: expected ETag {self.expected_etag}, got {self.actual_etag}"
+
+
+@dataclass
+class RetryConfig:
+    """Configuration for retry behavior on conflicts."""
+
+    max_retries: int = 5
+    base_delay: float = 0.1  # seconds
+    max_delay: float = 5.0  # seconds
+    jitter: float = 0.1  # random jitter factor
+
+
+class S3WorkspaceBackend:
+    """
+    S3 backend with optimistic concurrency control using conditional writes.
+
+    Uses:
+    - If-None-Match: * for create-if-not-exists operations
+    - If-Match: <etag> for update-only-if-unchanged operations
+    """
+
+    def __init__(
+        self,
+        bucket: str,
+        prefix: str = "",
+        s3_client: Any | None = None,
+        retry_config: RetryConfig | None = None,
+    ):
+        self.bucket = bucket
+        self.prefix = prefix.rstrip("/") + "/" if prefix else ""
+        self._client = s3_client
+        self.retry_config = retry_config or RetryConfig()
+        # Local cache of ETags for optimistic locking
+        self._etag_cache: dict[str, str] = {}
+
+    @property
+    def client(self) -> Any:
+        """Lazy initialization of S3 client."""
+        if self._client is None:
+            import boto3
+
+            self._client = boto3.client("s3")
+        return self._client
+
+    def _full_key(self, path: str) -> str:
+        """Convert a normalized path to a full S3 key."""
+        if path == ".":
+            return self.prefix.rstrip("/") if self.prefix else ""
+        return f"{self.prefix}{path}"
+
+    def _strip_prefix(self, key: str) -> str:
+        """Strip the prefix from an S3 key to get the relative path."""
+        if self.prefix and key.startswith(self.prefix):
+            return key[len(self.prefix) :]
+        return key
+
+    def _get_with_etag(self, path: str) -> tuple[str, str]:
+        """Read a file and return (content, etag)."""
+        key = self._full_key(path)
+        try:
+            response = self.client.get_object(Bucket=self.bucket, Key=key)
+            content = response["Body"].read().decode("utf-8")
+            etag = response["ETag"].strip('"')
+            self._etag_cache[key] = etag
+            return content, etag
+        except self.client.exceptions.NoSuchKey:
+            raise FileNotFoundError(f"{path} does not exist")
+
+    def _put_with_condition(
+        self,
+        path: str,
+        content: str,
+        *,
+        if_none_match: bool = False,
+        if_match: str | None = None,
+    ) -> str:
+        """
+        Write a file with conditional headers.
+
+        Args:
+            path: The file path
+            content: The content to write
+            if_none_match: If True, only write if file doesn't exist
+            if_match: If provided, only write if ETag matches
+
+        Returns:
+            The new ETag of the written object
+
+        Raises:
+            ConflictError: If the condition fails
+            FileExistsError: If if_none_match=True and file exists
+        """
+        key = self._full_key(path)
+        kwargs: dict[str, Any] = {
+            "Bucket": self.bucket,
+            "Key": key,
+            "Body": content.encode("utf-8"),
+            "ContentType": "text/plain; charset=utf-8",
+        }
+
+        if if_none_match:
+            kwargs["IfNoneMatch"] = "*"
+        elif if_match:
+            kwargs["IfMatch"] = (
+                f'"{if_match}"' if not if_match.startswith('"') else if_match
+            )
+
+        try:
+            response = self.client.put_object(**kwargs)
+            new_etag = response["ETag"].strip('"')
+            self._etag_cache[key] = new_etag
+            return new_etag
+        except self.client.exceptions.ClientError as e:
+            error_code = e.response.get("Error", {}).get("Code", "")
+            if error_code == "PreconditionFailed":
+                if if_none_match:
+                    raise FileExistsError(f"{path} already exists")
+                raise ConflictError(
+                    key=key,
+                    expected_etag=if_match,
+                    actual_etag=None,
+                    message=f"File {path} was modified by another process",
+                )
+            raise
+
+    def _retry_with_backoff(
+        self,
+        operation: str,
+        path: str,
+        func: Any,
+    ) -> Any:
+        """Execute a function with retry and exponential backoff on conflicts."""
+        config = self.retry_config
+        last_error: Exception | None = None
+
+        for attempt in range(config.max_retries + 1):
+            try:
+                return func()
+            except ConflictError as e:
+                last_error = e
+                if attempt >= config.max_retries:
+                    break
+                # Calculate backoff with jitter
+                delay = min(
+                    config.base_delay * (2**attempt),
+                    config.max_delay,
+                )
+                jitter = delay * config.jitter * random.random()
+                time.sleep(delay + jitter)
+                # Clear cached ETag to force fresh read
+                key = self._full_key(path)
+                self._etag_cache.pop(key, None)
+
+        raise last_error or RuntimeError(f"Retry failed for {operation} on {path}")
+
+    def read_file(self, path: str) -> str:
+        """Read a file from S3."""
+        key = _normalize_path(path)
+        content, _ = self._get_with_etag(key)
+        return content
+
+    def write_file(self, path: str, content: str, *, overwrite: bool) -> None:
+        """
+        Write a file to S3 with optimistic locking.
+
+        If overwrite=False, uses If-None-Match: * to ensure create-only.
+        If overwrite=True, uses If-Match with cached ETag if available,
+        otherwise does unconditional write.
+        """
+        key = _normalize_path(path)
+        s3_key = self._full_key(key)
+
+        if not overwrite:
+            # Create-if-not-exists
+            self._put_with_condition(key, content, if_none_match=True)
+        else:
+            # Check if we have a cached ETag for optimistic locking
+            cached_etag = self._etag_cache.get(s3_key)
+            if cached_etag:
+                # Use optimistic locking with retry
+                def _do_write():
+                    nonlocal cached_etag
+                    try:
+                        self._put_with_condition(key, content, if_match=cached_etag)
+                    except ConflictError:
+                        # Refresh ETag and retry
+                        _, cached_etag = self._get_with_etag(key)
+                        raise
+
+                self._retry_with_backoff("write_file", key, _do_write)
+            else:
+                # No cached ETag - try to get current state first for safety
+                try:
+                    _, etag = self._get_with_etag(key)
+                    self._put_with_condition(key, content, if_match=etag)
+                except FileNotFoundError:
+                    # File doesn't exist, create it
+                    self._put_with_condition(key, content, if_none_match=True)
+
+    def append_file(self, path: str, content: str) -> None:
+        """
+        Append to a file with optimistic locking.
+
+        Uses read-modify-write with If-Match for consistency.
+        """
+        key = _normalize_path(path)
+
+        def _do_append():
+            try:
+                current, etag = self._get_with_etag(key)
+                new_content = current + content
+            except FileNotFoundError:
+                # File doesn't exist, create it
+                self._put_with_condition(key, content, if_none_match=True)
+                return
+
+            self._put_with_condition(key, new_content, if_match=etag)
+
+        self._retry_with_backoff("append_file", key, _do_append)
+
+    def delete_path(self, path: str) -> None:
+        """Delete a file or directory from S3."""
+        key = _normalize_path(path, allow_root=True)
+
+        if key == ".":
+            # Delete all files under the prefix
+            paginator = self.client.get_paginator("list_objects_v2")
+            prefix = self.prefix if self.prefix else ""
+            for page in paginator.paginate(Bucket=self.bucket, Prefix=prefix):
+                for obj in page.get("Contents", []):
+                    self.client.delete_object(Bucket=self.bucket, Key=obj["Key"])
+                    self._etag_cache.pop(obj["Key"], None)
+            return
+
+        s3_key = self._full_key(key)
+
+        # Try to delete as a single file first
+        try:
+            self.client.head_object(Bucket=self.bucket, Key=s3_key)
+            self.client.delete_object(Bucket=self.bucket, Key=s3_key)
+            self._etag_cache.pop(s3_key, None)
+            return
+        except self.client.exceptions.ClientError:
+            pass
+
+        # Try as a directory prefix
+        prefix = f"{s3_key}/"
+        paginator = self.client.get_paginator("list_objects_v2")
+        deleted_any = False
+
+        for page in paginator.paginate(Bucket=self.bucket, Prefix=prefix):
+            for obj in page.get("Contents", []):
+                self.client.delete_object(Bucket=self.bucket, Key=obj["Key"])
+                self._etag_cache.pop(obj["Key"], None)
+                deleted_any = True
+
+        if not deleted_any:
+            raise FileNotFoundError(f"{key} does not exist")
+
+    def list_dir(self, path: str, recursive: bool) -> list[dict[str, Any]]:
+        """List directory contents from S3."""
+        key = _normalize_path(path, allow_root=True)
+
+        if key == ".":
+            prefix = self.prefix
+        else:
+            prefix = f"{self._full_key(key)}/"
+
+        entries: list[dict[str, Any]] = []
+        seen_dirs: set[str] = set()
+
+        paginator = self.client.get_paginator("list_objects_v2")
+        kwargs: dict[str, Any] = {"Bucket": self.bucket, "Prefix": prefix}
+
+        if not recursive:
+            kwargs["Delimiter"] = "/"
+
+        for page in paginator.paginate(**kwargs):
+            # Handle common prefixes (directories) in non-recursive mode
+            for common_prefix in page.get("CommonPrefixes", []):
+                dir_key = common_prefix["Prefix"].rstrip("/")
+                rel_path = self._strip_prefix(dir_key)
+                if rel_path and rel_path not in seen_dirs:
+                    entries.append(
+                        {"path": rel_path, "type": "directory", "size": None}
+                    )
+                    seen_dirs.add(rel_path)
+
+            # Handle files
+            for obj in page.get("Contents", []):
+                obj_key = obj["Key"]
+                rel_path = self._strip_prefix(obj_key)
+
+                # Skip the prefix itself if it's an empty marker
+                if not rel_path or rel_path.endswith("/"):
+                    continue
+
+                if recursive:
+                    entries.append(self._format_file_entry(rel_path, obj))
+                else:
+                    # Check if this is a direct child
+                    remainder = rel_path
+                    if key != ".":
+                        # Remove the directory prefix to get relative path
+                        dir_prefix = key + "/"
+                        if rel_path.startswith(dir_prefix):
+                            remainder = rel_path[len(dir_prefix) :]
+                        else:
+                            continue
+
+                    if "/" not in remainder:
+                        entries.append(self._format_file_entry(rel_path, obj))
+
+        # Sort by path for consistent ordering
+        entries.sort(key=lambda e: e["path"])
+
+        if not entries and key != ".":
+            # Check if the path itself exists as a file
+            try:
+                s3_key = self._full_key(key)
+                response = self.client.head_object(Bucket=self.bucket, Key=s3_key)
+                return [
+                    {
+                        "path": key,
+                        "type": "file",
+                        "size": response["ContentLength"],
+                        "etag": response["ETag"].strip('"'),
+                    }
+                ]
+            except self.client.exceptions.ClientError:
+                raise FileNotFoundError(f"{key} does not exist")
+
+        return entries
+
+    def grep(self, pattern: str, path: str | None, limit: int) -> list[dict[str, Any]]:
+        """Search for pattern in files."""
+        regex = re.compile(pattern)
+        key = _normalize_path(path, allow_root=True) if path is not None else "."
+
+        matches: list[dict[str, Any]] = []
+        paginator = self.client.get_paginator("list_objects_v2")
+
+        def process_object(obj_key: str, rel_path: str) -> None:
+            if len(matches) >= limit:
+                return
+            try:
+                response = self.client.get_object(Bucket=self.bucket, Key=obj_key)
+                content = response["Body"].read().decode("utf-8")
+
+                for line_no, line in enumerate(content.splitlines(), start=1):
+                    if regex.search(line):
+                        matches.append(
+                            {
+                                "path": rel_path,
+                                "line": line_no,
+                                "text": line.strip(),
+                            }
+                        )
+                        if len(matches) >= limit:
+                            return
+            except Exception:
+                # Skip files that can't be read as text
+                return
+
+        # If a specific path is provided, check if it's a file first
+        if key == ".":
+            prefix = self.prefix
+        else:
+            s3_key = self._full_key(key)
+            try:
+                self.client.head_object(Bucket=self.bucket, Key=s3_key)
+                rel_path = self._strip_prefix(s3_key)
+                process_object(s3_key, rel_path)
+                return matches
+            except self.client.exceptions.ClientError:
+                prefix = f"{s3_key}/"
+
+        for page in paginator.paginate(Bucket=self.bucket, Prefix=prefix):
+            for obj in page.get("Contents", []):
+                if len(matches) >= limit:
+                    return matches
+
+                obj_key = obj["Key"]
+                rel_path = self._strip_prefix(obj_key)
+
+                if not rel_path or rel_path.endswith("/"):
+                    continue
+
+                process_object(obj_key, rel_path)
+
+        return matches
+
+    def _format_file_entry(self, path: str, obj: dict[str, Any]) -> dict[str, Any]:
+        """Format a file entry from S3 object metadata."""
+        return {
+            "path": path,
+            "type": "file",
+            "size": obj["Size"],
+            "etag": obj["ETag"].strip('"'),
+        }
+
+    def get_file_etag(self, path: str) -> str | None:
+        """Get the cached ETag for a file, or fetch it if not cached."""
+        key = _normalize_path(path)
+        s3_key = self._full_key(key)
+
+        if s3_key in self._etag_cache:
+            return self._etag_cache[s3_key]
+
+        try:
+            response = self.client.head_object(Bucket=self.bucket, Key=s3_key)
+            etag = response["ETag"].strip('"')
+            self._etag_cache[s3_key] = etag
+            return etag
+        except self.client.exceptions.ClientError:
+            return None
+
+
+# Command types for the S3 filesystem tool
+S3FsCommand = Literal[
+    "read_file",
+    "write_file",
+    "delete_path",
+    "list_dir",
+    "grep",
+]
+
+ALL_S3_COMMANDS: tuple[S3FsCommand, ...] = (
+    "read_file",
+    "write_file",
+    "delete_path",
+    "list_dir",
+    "grep",
+)
+
+
+class S3FilesystemParams(BaseModel):
+    """Schema describing S3 filesystem tool calls."""
+
+    command: S3FsCommand = Field(
+        description="Filesystem operation to perform (read_file, write_file, delete_path, list_dir, grep)"
+    )
+    path: Optional[str] = Field(
+        default=None,
+        description="Path to operate on, relative to workspace root. Use '.' for the root directory.",
+    )
+    start_line: Optional[int] = Field(
+        default=None,
+        description="1-based inclusive start line when reading a file. Leave unset to read from the beginning.",
+        ge=1,
+    )
+    end_line: Optional[int] = Field(
+        default=None,
+        description="1-based inclusive end line when reading a file. Leave unset to read through the end.",
+        ge=1,
+    )
+    content: Optional[str] = Field(
+        default=None,
+        description="Content to write when using write_file.",
+    )
+    mode: Optional[Literal["overwrite", "append", "create_if_missing"]] = Field(
+        default="overwrite",
+        description="How to write content. Overwrite replaces the file, append adds to the end, create_if_missing leaves existing files untouched.",
+    )
+    recursive: Optional[bool] = Field(
+        default=None,
+        description="When listing directories, set to true to recurse.",
+    )
+    pattern: Optional[str] = Field(
+        default=None,
+        description="Regular expression pattern to search for when using grep.",
+    )
+    max_results: Optional[int] = Field(
+        default=50,
+        description="Maximum number of grep matches to return.",
+        ge=1,
+    )
+
+
+class S3FilesystemManager:
+    """
+    S3-backed filesystem manager with optimistic concurrency control.
+
+    Uses S3 conditional writes (If-None-Match and If-Match) for safe distributed
+    operations, allowing multiple AI agents to share filesystem state.
+
+    Example:
+        manager = S3FilesystemManager(
+            bucket="my-ai-workspace",
+            prefix="agent-123/",
+        )
+        tools = manager.get_tools()
+    """
+
+    def __init__(
+        self,
+        bucket: str,
+        prefix: str = "",
+        s3_client: Any | None = None,
+        retry_config: RetryConfig | None = None,
+        tool_name: str = "s3_filesystem",
+    ):
+        """
+        Initialize the S3 filesystem manager.
+
+        Args:
+            bucket: The S3 bucket name
+            prefix: Optional prefix for all keys (like a workspace directory)
+            s3_client: Optional pre-configured S3 client
+            retry_config: Configuration for retry behavior on conflicts
+            tool_name: Name for the tool (default: "s3_filesystem")
+        """
+        self.backend = S3WorkspaceBackend(
+            bucket=bucket,
+            prefix=prefix,
+            s3_client=s3_client,
+            retry_config=retry_config,
+        )
+        self.tool_name = tool_name
+        self._tool_cache: dict[tuple[str, ...], list[Tool]] = {}
+
+    def _handle_read(
+        self, path: str, start_line: Optional[int], end_line: Optional[int]
+    ) -> dict[str, Any]:
+        content = self.backend.read_file(path)
+        total_lines = len(content.splitlines()) or (0 if content == "" else 1)
+        start = start_line or 1
+        end = end_line or total_lines
+        if end < start:
+            if not (total_lines == 0 and end_line is None and start == 1):
+                raise ValueError("end_line must be greater than or equal to start_line")
+
+        if start == 1 and end >= total_lines:
+            snippet = content
+        else:
+            lines = content.splitlines()
+            snippet = "\n".join(lines[start - 1 : end])
+
+        # Include ETag in response for transparency
+        etag = self.backend.get_file_etag(path)
+
+        return {
+            "path": path,
+            "start_line": start,
+            "end_line": end,
+            "content": snippet,
+            "total_lines": total_lines,
+            "character_count": len(content),
+            "etag": etag,
+        }
+
+    def _handle_write(
+        self, path: str, content: str, mode: Optional[str]
+    ) -> dict[str, Any]:
+        write_mode = mode or "overwrite"
+        if write_mode == "overwrite":
+            self.backend.write_file(path, content, overwrite=True)
+        elif write_mode == "append":
+            self.backend.append_file(path, content)
+        elif write_mode == "create_if_missing":
+            try:
+                self.backend.write_file(path, content, overwrite=False)
+            except FileExistsError:
+                pass
+        else:
+            raise ValueError(f"Unsupported write mode: {write_mode}")
+
+        # Get new ETag after write
+        etag = self.backend.get_file_etag(path)
+
+        return {"path": path, "status": "ok", "mode": write_mode, "etag": etag}
+
+    def _handle_delete(self, path: str) -> dict[str, Any]:
+        self.backend.delete_path(path)
+        return {"path": path, "status": "ok"}
+
+    def _handle_list(
+        self, path: Optional[str], recursive: Optional[bool]
+    ) -> dict[str, Any]:
+        listing = self.backend.list_dir(path or ".", recursive=bool(recursive))
+        return {"path": path or ".", "recursive": bool(recursive), "entries": listing}
+
+    def _handle_grep(
+        self, pattern: str, path: Optional[str], limit: Optional[int]
+    ) -> dict[str, Any]:
+        max_results = limit or 50
+        matches = self.backend.grep(pattern, path=path, limit=max_results)
+        return {
+            "pattern": pattern,
+            "path": path,
+            "max_results": max_results,
+            "matches": matches,
+        }
+
+    def _filesystem_tool(self, allowed_commands: set[str], **kwargs: Any) -> str:
+        params = S3FilesystemParams.model_validate(kwargs)
+
+        try:
+            if params.command not in allowed_commands:
+                raise ValueError(
+                    f"The '{params.command}' command is disabled for this tool instance"
+                )
+            if params.command == "read_file":
+                if not params.path:
+                    raise ValueError("path is required for read_file")
+                result = self._handle_read(
+                    params.path, params.start_line, params.end_line
+                )
+            elif params.command == "write_file":
+                if params.path is None or params.content is None:
+                    raise ValueError("path and content are required for write_file")
+                result = self._handle_write(params.path, params.content, params.mode)
+            elif params.command == "delete_path":
+                if not params.path:
+                    raise ValueError("path is required for delete_path")
+                result = self._handle_delete(params.path)
+            elif params.command == "list_dir":
+                result = self._handle_list(params.path, params.recursive)
+            elif params.command == "grep":
+                if not params.pattern:
+                    raise ValueError("pattern is required for grep")
+                result = self._handle_grep(
+                    params.pattern, params.path, params.max_results
+                )
+            else:
+                raise ValueError(f"Unknown command: {params.command}")
+            return json.dumps({"ok": True, "result": result}, indent=2)
+        except Exception as exc:
+            return json.dumps(
+                {"ok": False, "error": type(exc).__name__, "message": str(exc)},
+                indent=2,
+            )
+
+    def get_tools(self, *, exclude: Iterable[S3FsCommand] | None = None) -> list[Tool]:
+        """
+        Get the filesystem tools.
+
+        Args:
+            exclude: Optional list of commands to exclude from the tool
+
+        Returns:
+            List containing the S3 filesystem tool
+        """
+        exclude_set = set(exclude or [])
+        unknown = exclude_set.difference(ALL_S3_COMMANDS)
+        if unknown:
+            raise ValueError(f"Unknown commands in exclude list: {sorted(unknown)}")
+
+        allowed = tuple(cmd for cmd in ALL_S3_COMMANDS if cmd not in exclude_set)
+        if not allowed:
+            raise ValueError("Cannot exclude every filesystem command")
+
+        cache_key = allowed
+        if cache_key in self._tool_cache:
+            return self._tool_cache[cache_key]
+
+        allowed_set = {cmd for cmd in allowed}
+        schema = S3FilesystemParams.model_json_schema(ref_template="#/$defs/{model}")
+        if (
+            "properties" in schema
+            and "command" in schema["properties"]
+            and isinstance(schema["properties"]["command"], dict)
+        ):
+            schema["properties"]["command"]["enum"] = list(allowed)
+
+        tool = Tool(
+            name=self.tool_name,
+            description=S3_FS_DESCRIPTION,
+            parameters=schema.get("properties", {}),
+            required=schema.get("required", []),
+            definitions=schema.get("$defs"),
+            run=partial(self._filesystem_tool, allowed_set),
+        )
+
+        self._tool_cache[cache_key] = [tool]
+        return [tool]
+
+
+__all__ = [
+    "FilesystemManager",
+    "FilesystemParams",
+    "InMemoryWorkspaceBackend",
+    "WorkspaceBackend",
+    "S3FilesystemManager",
+    "S3FilesystemParams",
+    "S3WorkspaceBackend",
+    "ConflictError",
+    "RetryConfig",
+]
+
+
+description = """
+S3-backed remote filesystem tool with optimistic concurrency control.
+
+Uses S3 conditional writes (If-None-Match and If-Match) for safe distributed
+operations, allowing multiple AI agents to share filesystem state without conflicts.
+
+Features:
+- If-None-Match: * -> Create-if-not-exists (distributed locks, idempotent writes)
+- If-Match: <etag> -> Update-only-if-unchanged (optimistic locking)
+- Automatic retry with exponential backoff on conflicts
+- ETag tracking for all file operations
+
+Example:
+    from lm_deluge.tool.prefab.s3_filesystem import S3FilesystemManager
+
+    manager = S3FilesystemManager(
+        bucket="my-ai-workspace",
+        prefix="agent-123/",  # Optional: isolate agent's workspace
+    )
+
+    # Get tools for the agent
+    tools = manager.get_tools()
+
+    # The filesystem operations are now safe for concurrent access
+"""