genius-evolve
Use when the user runs /genius-evolve or asks to run the EVOLVE engine on an already-prepared evolve-dir (initial_program.py + frozen evaluator.py + frozen.lock).
Datos de origen
- Repositorio
- chenpg2/genius-plugin
- Última actividad en el origen
- 24 de junio de 2026 a las 04:51
- Idioma detectado de SKILL.md
- inglés
- Estrellas
- 3
- Forks
- 0
Opciones de instalación
De forma predeterminada está seleccionado el prompt que primero revisa el origen. Puedes cambiar a un comando directo o descargar una copia local.
Revisa los archivos de origen
Lee SKILL.md y los archivos complementarios que muestra SkillsMP antes de decidir si quieres instalarlo.
Mostrando SKILL.md
SKILL.md
Instrucciones de origen · Vista previa de solo lectura- name
- genius-evolve
- description
- Use when the user runs /genius-evolve or asks to run the EVOLVE engine on an already-prepared evolve-dir (initial_program.py + frozen evaluator.py + frozen.lock).
# GENIUS — evolve (standalone)
Preconditions: the evolve-dir already contains `initial_program.py`, `evaluator.py`, and a
`frozen.lock` (sha256 of the evaluator). If not, first do the draft→freeze gate from the `genius`
skill's EVOLVE section. Setup: see `$CLAUDE_PLUGIN_ROOT/docs/EVOLVE-SETUP.md` (ollama + the
openevolve venv).
1. Resolve `PLUGIN_ROOT` (`echo "$CLAUDE_PLUGIN_ROOT"`).
2. **Verify preconditions — fail explicitly, no silent fallback.** Confirm `<evolve-dir>/initial_program.py`,
`<evolve-dir>/evaluator.py`, and `<evolve-dir>/frozen.lock` all exist (e.g. `ls -1 <evolve-dir>/{initial_program.py,evaluator.py,frozen.lock}`).
If any is missing, report exactly which one(s) and STOP — direct the user to the draft→freeze gate
in the `genius` skill's EVOLVE section. Do NOT attempt to run EVOLVE on an under-prepared dir.
3. Run: `bash "$CLAUDE_PLUGIN_ROOT/scripts/run-openevolve.sh" <evolve-dir> --iterations <N>`. Use the
user's `--iterations N` override from the command args if present, else default N=30. On non-zero
exit, report the reason (frozen mismatch / missing venv / no best produced).
4. On success, read `<evolve-dir>/out/best/best_program_info.json` and report the evolved artifact +
its `combined_score` + the trajectory.
Ver en GitHub