Skip to content

Security: MinhMPA/claude-code-local-reader

Security

docs/security.md

Security

Security model

This project assumes:

  • Claude Code remains the main agent.
  • The local model is used only as a read-only reader.
  • The intended endpoint is a localhost LM Studio server.

What read-only means here

The local reader can summarize, compress, and extract context. It does not directly edit files, apply patches, run git operations, or make final correctness decisions in this workflow.

What is still exposed

Anything you send through ask-local, qdoc, qlog, qdiff, qgrep, or qapi is visible to the local model serving LM Studio. Treat that as data disclosure to software running on your machine.

Do not pass:

  • secrets
  • production credentials
  • sensitive customer data
  • logs with private tokens or internal identifiers

unless you trust the machine, the local environment, and the model you loaded.

Public sharing notes

This repo is designed to be safe to publish because it:

  • uses <your-model-id> instead of a personal model id
  • does not require private paths
  • does not include credentials
  • defaults to http://127.0.0.1:1234/v1

Before sharing your own configs or shell history, still check for local secrets or machine-specific paths.

Network exposure warning

If you expose LM Studio beyond localhost, the trust boundary changes. This repo does not assume or recommend that. v1 is a localhost-first workflow.

There aren't any published security advisories