filesystem-tool
Skill0 starsChanged 6 months agoMITView on GitHub
Implements a sandboxed file system tool for agents using @tool() class-form. Use when building tools that read, write, or list files within a restricted base directory, with path traversal protection to prevent access outside the allowed workspace.
In this repository65 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/building-agents/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- llms-full.txtllms.txt
- llms.txtllms.txt
- adding-guardrailsSkill
- agent-health-timeoutSkill
- agent-permission-scopingSkill
- agent-session-isolationSkill
- agent-step-loggingSkill
- agent-supervisor-routerSkill
- agent-telemetrySkill
- async-agent-backgroundSkill
- building-agentsSkill
- building-teamsSkill
- building-toolsSkill
- calendar-scheduling-toolSkill
- callback-tracing-handlerSkill
- chain-of-thought-promptingSkill
- chat-model-configurationSkill
- code-execution-sandboxSkill
- common-agent-patternsSkill
- context-window-managementSkill
- conversation-memorySkill
- conversation-state-dbSkill
- database-query-toolSkill
- document-loadersSkill
- document-rerankingSkill
- email-notification-toolSkill
- embedding-model-ingestionSkill
- fallback-model-chainSkill
- few-shot-injectionSkill
- human-in-the-loopSkill
- hybrid-searchSkill
- inspecting-streamsSkill
- integrating-with-laurenSkill
- json-schema-enforcementSkill
- llm-provider-connectionSkill
- llm-rate-limitingSkill
- long-term-vector-memorySkill
- managing-memorySkill
- migrating-to-lauren-aiSkill
- multi-agent-delegationSkill
- multi-tool-orchestrationSkill
- multi-turn-tool-stateSkill
- output-guardrailsSkill
- plan-and-execute-agentSkill
- prompt-injection-defenseSkill
- prompt-versioningSkill
- rag-pipelineSkill
- react-agent-loopSkill
- response-validation-retrySkill
- rest-api-toolSkill
- securing-agentsSkill
- semantic-cacheSkill
- streaming-response-handlerSkill
- structured-output-parsingSkill
- sub-agent-spawningSkill
- system-prompt-templatingSkill
- testing-agentsSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 82
Use tempfile.TemporaryDirectory for isolated test directories.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Use
codemap find "SymbolName"to locate any symbol before reading — it gives exact file + line range and is faster than grep across the whole repo.
File System Operation Tool with Path Restrictions
Critical rule — tool annotations must resolve
from __future__ import annotations is supported, but every type used by @tool() must resolve when schema generation runs.
@tool() resolves annotations when it builds the JSON schema. Future
annotations are supported, but unresolved forward references and circular
imports in function-form tool files still break schema generation.
Overview
The FileSystemTool is a class-form @tool() that reads, writes, and lists
files within a configurable base directory. It resolves all paths via
Path.resolve() and rejects any path that resolves outside the base — blocking
../ traversal attacks.
Implementation
# tools/filesystem_tool.py — future annotations are allowed; keep tool types importable
from pathlib import Path
from lauren_ai import tool, ToolContext
@tool()
class FileSystemTool:
"""Read and write files within the allowed directory.
Args:
operation: 'read', 'write', or 'list'.
path: Relative path within the allowed directory.
content: Content to write (for write operation).
"""
def __init__(self, base_path: str = "/tmp/agent-workspace"):
self._base = Path(base_path).resolve()
self._base.mkdir(parents=True, exist_ok=True)
def _safe_path(self, relative: str) -> Path | None:
target = (self._base / relative).resolve()
if not str(target).startswith(str(self._base)):
return None # path traversal detected
return target
async def run(
self, ctx: ToolContext, operation: str, path: str, content: str = ""
) -> dict:
safe = self._safe_path(path)
if safe is None:
return {"error": "Access denied: path traversal detected"}
if operation == "read":
if not safe.exists():
return {"error": f"File not found: {path}"}
return {"content": safe.read_text(encoding="utf-8")}
elif operation == "write":
safe.parent.mkdir(parents=True, exist_ok=True)
safe.write_text(content, encoding="utf-8")
return {"written": str(path), "bytes": len(content)}
elif operation == "list":
if not safe.exists() or not safe.is_dir():
return {"files": []}
return {"files": [f.name for f in safe.iterdir()]}
return {"error": f"Unknown operation: {operation}"}AgentRunner test pattern
Use tempfile.TemporaryDirectory for isolated test directories. Create a fresh
tool instance and agent factory per test to avoid shared state.
import json
import tempfile
from lauren_ai._agents import AgentContext, agent, use_tools
from lauren_ai._tools import ToolResult
from lauren_ai._transport import Completion, TokenUsage
from lauren_ai.testing import TestClient
class _Capture:
def __init__(self):
self.captured: list[ToolResult] = []
async def on_tool_result(self, result: ToolResult, ctx: AgentContext) -> ToolResult | None:
self.captured.append(result)
return None
def _make_agent(base_path: str):
fs_tool = FileSystemTool(base_path=base_path)
@agent(model=None, system="Filesystem agent")
@use_tools(fs_tool)
class FsTestAgent(_Capture):
def __init__(self):
_Capture.__init__(self)
return FsTestAgent()
def _c(text):
return Completion(id="c1", model="mock", content=text, tool_calls=[],
stop_reason="end_turn", usage=TokenUsage(10, 5))
def test_write_and_read():
with tempfile.TemporaryDirectory() as tmpdir:
agent_inst = _make_agent(tmpdir)
client = TestClient(agent_inst)
client.mock.queue_tool_use(
"file_system_tool",
{"operation": "write", "path": "test.txt", "content": "hello"},
)
client.mock.queue_response(_c("Done."))
client.run("Write test.txt")
out = json.loads(agent_inst.captured[0].content)
assert out["written"] == "test.txt"
def test_path_traversal_blocked():
with tempfile.TemporaryDirectory() as tmpdir:
agent_inst = _make_agent(tmpdir)
client = TestClient(agent_inst)
client.mock.queue_tool_use(
"file_system_tool",
{"operation": "read", "path": "../../etc/passwd"},
)
client.mock.queue_response(_c("Access denied."))
client.run("Try to read /etc/passwd")
out = json.loads(agent_inst.captured[0].content)
assert "error" in out
assert "traversal" in out["error"].lower() or "denied" in out["error"].lower()Path safety rules
| Input path | Result |
"notes.txt" | Allowed — stays in base |
"subdir/notes.txt" | Allowed — stays in base |
"../../etc/passwd" | Blocked — resolves outside base |
"/etc/passwd" | Blocked — absolute path outside base |
The safety check uses str(target).startswith(str(self._base)) after both
paths are fully resolved. This is robust against symlink tricks on most
platforms.
Reference files
| File | Contents |
src/lauren_ai/_tools/__init__.py | @tool(), ToolContext |
src/lauren_ai/_tools/_executor.py | ToolExecutor dispatch |
src/lauren_ai/_tools/_schema.py | JSON schema generation |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

