rustport

Port games and apps to Rust, including compatible game runtimes that load the original maps, assets, scripts and saves. Capture the original's behavior, port a small playable slice, then measure compatibility as the project grows.

One Agent Skills bundle for Codex, Claude Code, OpenCode, Pi, DeepSeek Harness (DSH), and Hermes. No model or MCP setup is changed by the installer.

Install

Requires Python 3.9+. Clone first:

git clone https://github.com/LakoMoor/rustport-skill.git
cd rustport-skill
python3 install.py --agent all

Or choose agents and scope:

python3 install.py --agent codex --agent claude
python3 install.py --agent opencode --project
python3 install.py --agent pi --agent dsh --agent hermes --project
python3 install.py --agent all --dry-run

./install.sh remains a Bash entrypoint; without options it installs for Claude Code, preserving the previous behavior. On Windows use py -3 install.py instead of python3. The bundled shell helpers need Bash (Git Bash/WSL), Python 3 and Cargo. GPU/platform support depends on the target and the wgpu toolchain.

Existing differing files are preserved unless you explicitly pass --force. --root PATH selects a project root; --home PATH selects an alternate user home. See agent locations and environment overrides.

Invoke

Agent	Explicit command
Codex	/skills → choose rustport, or $rustport <request>
Claude Code	/rustport <request>
OpenCode	/rustport <request>
Pi	/rustport <request>; also /skill:rustport <request>
DeepSeek Harness	/rustport <request>
Hermes	/rustport <request>

OpenCode gets a command file; Pi gets a prompt template. Both load the same skill bundle and pass your arguments. Other agents use native skill discovery. Codex uses its native skill selector rather than an unsupported /rustport alias. Reload/restart after installing if needed; Pi has /reload. Hermes project skills require project trust (hermes skills trust after reviewing the bundle). DSH needs its filesystem skill provider enabled. Host command/trust settings still apply.

Examples (use the invocation syntax for your agent):

/rustport Port ~/games/snake from C++/SDL2 to Rust with wgpu
/rustport Reimplement this game's runtime in Rust, loading my original maps and scripts
/rustport Continue this port: fix save/load compatibility and compare with the original

What the agent does

• Source port: survey → scope → original-side golden tests → Rust core → rendering → measured parity.
• Compatible runtime: pin the reference build, inspect real content, maintain a compatibility ledger, and bring up a playable slice through the real game loop.
• Continued work: resume the ledger, reproduce a behavior gap, repair the actual runtime path, and update evidence instead of restarting the project.

For Horizon/openOMSI-style work, the additional compatible-runtime workflow covers original formats and scripts, reference oracles, save/load, streaming, deterministic scenarios, agent control, and progress across sessions. This is a development workflow, not a promise that a complete game can be recreated in a single run.

Bundle

Path	Purpose
SKILL.md	Agent-neutral entrypoint and mode selection
references/agents.md	Agent discovery, commands and upstream documentation
references/compatible-runtime.md	Whole-game runtime reconstruction
scripts/survey.sh	Source inventory and risk hints
scripts/scaffold.sh	Headless core + Rust app workspace
scripts/parity.sh	Golden output comparisons
assets/wgpu_template/	winit 0.30 + wgpu 30 starter
assets/mappings.md	Language/library map and graphics notes
assets/PORT_TEMPLATE.md	Living port plan

With only a shipped binary, APK or Electron app, use available reverse-engineering and runtime-observation tools. REA integration is optional; the skill must disclose unavailable tools and unverified behavior rather than invent results. Original game content stays local to the user's installation.

Workbench: evidence-led Rust ports

The optional Python standard-library workbench augments the agent skill with a durable SQLite function/behavior ledger, executable SHA-256 pinning, strict JSON/JSONL trace comparisons, stale-proof audits and opt-in dual-process capture. It does not automatically decompile, execute or certify a whole game.

python3 scripts/workbench.py doctor
python3 scripts/workbench.py pin --original /path/to/Game.exe --output reference.lock.json
python3 scripts/workbench.py ledger --db port-ledger.sqlite add --id input.step --origin 'Game.exe:0x1234' --target 'input::step' --evidence 'docs/TRACES.md'
python3 scripts/workbench.py ledger --db port-ledger.sqlite set --id input.step --status implemented --wired 1
python3 scripts/oracle_runner.py scenario.json --output ./capture --allow-execute-original --compare
python3 scripts/workbench.py parity ./capture/original.jsonl ./capture/port.jsonl --db port-ledger.sqlite --id input.step --reference-lock reference.lock.json
python3 scripts/workbench.py ledger --db port-ledger.sqlite audit

Read Workbench, reverse engineering routes, agent handoffs, scenario schema and the site workbench. The runner does not sandbox unsafe games; use an isolated VM for untrusted software. compared and verified are bounded trace statuses, not full-engine certification.

여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗