Native Rust Target Agent Execution Strategy (Replace Python Bridge)

[micahstubbs]

Goal: Replace the current Python subprocess bridge for executing target_agent.py with a native Rust solution.

This is now a high-priority item following the deprecation decision in #137.

Why this matters

• Full end-to-end native Rust self-improving loop.
• Removes external Python dependency for core functionality.
• Enables stronger safety guarantees and performance.
• Critical for Track 1 (Improve the Harness) credibility.

Options to Evaluate (in priority order for hackathon speed)

1. Embedded scripting (Rhai or Lua via mlua) — Fastest to implement, good for dynamic agent code.
2. WebAssembly (wasmtime + WASI) — Strongest safety model, aligns with capability-based security (Scheduler decision is observational only — make it actually drive the generation loop #90).
3. On-the-fly Rust compilation — Highest performance but slow iteration (not ideal for many generations).
4. Hybrid — Keep minimal Python for very complex target tasks during transition, but make native path the default.

Tasks

• Research and prototype the chosen approach.
• Define how tools, file I/O, and evaluation are exposed to the native target agent.
• Ensure trajectory capture remains high-fidelity.
• Update orchestrator to dispatch to native executor.
• Deprecate and eventually remove Python bridge code path.

Related: #137 (Deprecation decision), #62 (LLM runners), #84 (closed loop), #90 (sandboxing).

[micahstubbs]

Ready for Claude Opus 4.8 worker.

Claiming instructions: Comment "Claiming #138" below if you are starting work on this. This is currently the highest-priority remaining technical task after the deprecation decision (#137). Focus on designing a clean native Rust path for Target Agent execution (WASM, Rhai, or on-the-fly compilation) to fully replace the Python bridge.