# Drive Ida With Hexmux

> Drive running IDA Pro GUI instances through Hexmux by selecting the correct database, generating focused IDAPython scripts, executing them through the Hexmux CLI, and interpreting structured results. Use for reverse engineering, disassembly, decompilation, database inspection, renaming, comments, type changes, patching, or other IDAPython work when Hexmux is available.

- Skill: `chichou/drive-ida-with-hexmux` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add chichou/drive-ida-with-hexmux`
- Raw SKILL.md: https://api.skillmd.com/api/skills/chichou/drive-ida-with-hexmux/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: chichou (https://skillmd.com/u/chichou)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/chichou/drive-ida-with-hexmux

---


# Drive IDA with Hexmux

Use Hexmux as a small execution substrate, not as a catalog of narrow tools. Generate a task-specific script using verified IDAPython APIs and return a compact JSON-compatible `result`.

## Workflow

1. Run `hexmux ps --json`.
2. Select the instance by matching `idb_path`, `binary`, and PID to the user's target. If multiple entries remain plausible, ask the user; never silently use the first instance.
3. Look up uncertain IDAPython APIs in the available local IDAPython documentation. Read [references/idapython.md](references/idapython.md) for search locations and high-value modules.
4. Write one focused Python script. Prefer a saved `.py` file for nontrivial work so the user can inspect and rerun it.
5. Set a JSON-compatible global named `result`. Keep results small by filtering, limiting, and summarizing inside IDA.
6. Execute `hexmux run <instance-id-or-prefix> --json script.py`.
7. Check all of `ok`, `success`, `stderr`, and `traceback`. Correct the script based on actual errors; do not guess API names.

## Script conventions

Import only the modules needed by the script. Hexmux supplies a normal IDAPython execution namespace, not pre-imported convenience globals.

```python
import idautils
import ida_name

result = [
    {"ea": hex(ea), "name": ida_name.get_name(ea)}
    for ea in list(idautils.Functions())[:50]
]
```

Use strings for addresses when returning JSON to avoid ambiguity. For large text such as decompilation, return only the requested functions or write an artifact in the user's workspace when appropriate.

## Safety and retries

- Treat generated scripts as arbitrary code running inside IDA with the user's privileges.
- Inspect before executing mutations. Scope rename, comment, type, patch, debugger, and save operations exactly to the request.
- Do not retry a mutating script after a timeout or disconnect until inspecting IDA state. The client timeout cannot preempt Python already running on IDA's main thread.
- Avoid unbounded database walks. Add limits or incremental queries first, then expand with evidence.
- Prefer IDA APIs over parsing rendered UI text.

## Instance behavior

Each IDA process listens on a private Unix-domain socket named by its PID. `hexmux ps` discovers and queries those sockets directly. `hexmux run` requires an explicit PID or unique PID prefix and rejects ambiguous prefixes. If an instance is busy, wait for or investigate the active operation rather than sending competing mutations. Windows is not supported.