@tindalosko/dsh-cheatengine
English | 简体中文
A DeepSeek Harness (DSH) plugin that lets agents drive Cheat Engine for single-player game debugging: find values, find base addresses, lock resources, analyze writers, and more.
> ⚠️ Token warning: this plugin registers many ce_* tools, which increases token usage per request. Only ce_status, ce_connect, ce_tool_search, ce_playbook, and ce_mission are exposed by default; unlock tools on demand.
Scope
This plugin targets single-user / low-concurrency local single-player debugging: attach one target process at a time and debug from one DSH session. When multiple sessions debug concurrently, state and the floating panel follow the most recently active session; this plugin is not recommended for high-concurrency or multi-agent CE debugging.
Features
- Memory scan / filter / read / write
- Disassembly, breakpoints, registers, find-what-writes
- Pointer scan and base-address verification
- Address locking (infinite resources)
- AOB search / generation, module dump, speedhack, cheat table save/load
- Anti-cheat / protection module detection
- Session stats, hypothesis / evidence, audit / undo, snapshot, risk levels
- Unified memory / session tools (
ce_memory_read,ce_memory_write,ce_session); old tool names remain for compatibility - Optional floating status panel (bottom-right, closable / reopenable)
Quick start
1. CE side (Windows)
1. Get ce_mcp_bridge.lua and ce_mcp_tcp_x64.dll (use x86 for 32-bit CE) from cheatengine-mcp-tcp-bridge. 2. Put the DLL into your Cheat Engine directory. 3. Open CE and attach to the target process. 4. Run ce_mcp_bridge.lua; you should see Bridge started on port 17171.
You can also ask the DSH agent to call install_ce_bridge to automate steps 1–2.
2. DSH side
Install from npm (recommended):
dsh plugin add @tindalosko/dsh-cheatengineOr install from GitHub:
dsh plugin add github:TindalosKorone/dsh-cheatengineOr inject locally:
dev_inject_plugin {"dir": "/absolute/path/dsh-cheatengine"}Default bridge endpoint is 127.0.0.1:17171; override with ce_connect.
Usage
1. Run ce_status / ce_connect to verify the connection. 2. Use ce_tool_search to find and unlock tools by task pack, e.g. ce_tool_search({"packs": ["scan", "memory"]}); prefer ce_memory_read / ce_memory_write for memory access. 3. For common debugging flows, ask the agent to call ce_playbook / ce_mission.
See [AGENTS.md](AGENTS.md) for the full tool list and agent-facing conventions.
Floating panel
- Enabled by default; shows phase, calls, scan count, locks, and summary at the bottom-right.
- Click × to close; it becomes a small 🧊 CE button that reopens the panel.
- The panel only reads the local
/ce-status/api; it does not consume LLM tokens by itself (the plugin tools do).
Build & self-check
The repo ships a ready-to-run lib/, so cloning is enough. To build from source:
npm run build:all # build host + client
npm run typecheck
node scripts/self-check.mjs
node --test test/tools.test.mjsSafety
- Use only in environments where you have permission to debug the target.
- Dangerous tools modify memory or run scripts; think before unlocking.
- The bridge binds to local
127.0.0.1:17171by default; do not use over untrusted networks.
Links
- [AGENTS.md](AGENTS.md) — agent-facing usage guide
- HollyZoe/cheatengine-mcp-tcp-bridge — CE-side bridge