langchain-ai / langchain-ai/openshell-deepagent

agents.md file is somehow only visible through filesystem tools but not through python

Open
#2 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
191
Forks
33
PR merge metrics
No merged PRs in 30d

Description

There is something a bit strange going on with the CompositeBackend.

For example, if you ask for the contents of the Agents.md file, and if the LLM decides to use read_file then it can actually find the file. However, if you ask it to do so through a python script, so that it writes something like this:

import os

  os.makedirs('/sandbox', exist_ok=True)
  path = '/sandbox/memory/AGENTS.md'
  print('path:', path)
  print('exists:', os.path.exists(path))
  print('readable:', os.access(path, os.R_OK))
  try:
      with open(path, 'r', encoding='utf-8') as f:
          print(f.read())
  except Exception as e:
      print('error:', type(e).__name__, str(e))

It fails for some reason.. it seems the CompositeBackend doesn't work through Python!

The LLM eventually replies with..

I wrote and ran a Python script to read /sandbox/memory/AGENTS.md.
Result:

exists: False

readable: False

FileNotFoundError


So the same pattern holds:

filesystem tools can see /sandbox/memory/AGENTS.md

Python execution cannot access it


This suggests the execution environment and filesystem tools are seeing different mounted views of /sandbox.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by tracing the CompositeBackend behavior for filesystem tools and Python execution, using the reported /sandbox/memory/AGENTS.md example as the reproduction. Compare the filesystem views available to each path and define done as making the file consistently visible, or documenting the intended boundary with a regression test.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
backend, infrastructure
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.