← all repositories
aimeoa/hanshuang-codex

Desktop app rewriting system prompts for Cursor, Claude, and six others

Automates the messy file-system and cloud-memory work needed to inject custom system prompts into eight popular AI coding assistants.

hanshuang-codex
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does

This is an Electron desktop app with a React frontend that packages a suite of PowerShell scripts. Each script targets a specific AI assistant—Codex, Cursor, Claude, ZCode, WorkBuddy, DeepSeek Harness, or Doubao—and overwrites its local configuration, global memory, or system instructions to slip in custom prompts. For Doubao, the app goes further: it restarts the client in debug mode and uses Chrome DevTools Protocol to push memory to the cloud via the user’s own authenticated session.

The interesting bit

The project is essentially a configuration-management tool masquerading as a jailbreak utility. It maintains an exact map of where each assistant hides its system-level instructions, from ~/.codex/config.toml to Cursor’s User Rules and Claude’s CLAUDE.md. The Doubao target is the most unusual, hijacking the desktop client’s CDP session to write global memory remotely rather than touching local disk.

Key highlights

  • Targets eight distinct AI coding assistants with dedicated per-tool PowerShell injectors
  • Bundles its own UI design system (ported from PI-Desktop) and ships as Windows installer, portable EXE, or ZIP
  • Knows the exact injection point for each target: Codex config, Cursor global rules, Claude skills, WorkBuddy cloud memory blocks, DeepSeek Harness AGENTS.md, etc.
  • For Doubao, restarts the client in debug mode and writes global memory via CDP using the existing login session
  • Includes uninstall scripts that restore original configs; DeepSeek Harness backups are stored in ~/.dsh/managed-prompts/

Caveats

  • Windows-only: the build pipeline relies on NSIS, rcedit, and PowerShell scripts that target Windows paths and %APPDATA%
  • The repository metadata lists Python and PySide6, but the actual stack shown in the README is Electron + React + Node/npm; the mismatch is unexplained
  • The author explicitly warns that maintenance is sporadic due to day-job constraints, so updates are unpredictable

Verdict

Worth a look if you are researching how AI assistants store and load system-level instructions, or if you need to batch-modify prompts across multiple tools. Skip it if you are not on Windows, uncomfortable with prompt injection, or looking for an actively maintained project with guaranteed support.

Frequently asked

What is aimeoa/hanshuang-codex?
Automates the messy file-system and cloud-memory work needed to inject custom system prompts into eight popular AI coding assistants.
Is hanshuang-codex open source?
Yes — aimeoa/hanshuang-codex is open source, released under the MIT license.
What language is hanshuang-codex written in?
aimeoa/hanshuang-codex is primarily written in Python.
How popular is hanshuang-codex?
aimeoa/hanshuang-codex has 510 stars on GitHub.
Where can I find hanshuang-codex?
aimeoa/hanshuang-codex is on GitHub at https://github.com/aimeoa/hanshuang-codex.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.