计划执行与检查节点
executing-plans
按既定实施计划完成工作,保持过程检查和反馈。
msd_recovery_protocol【效率与协作】Assemble executable soft-mode recovery evidence for Minimum Stein Discrepancy Estimator proxy experiments.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Use this skill when recovering Minimum Stein Discrepancy Estimators under bounded runtime. It coordinates the generated Stein-kernel, diffusion-design, and optimiser skills to produce validation-ready recovery artifacts without reading an original source repository.
module_plan.json, module docs, generated skills, and environment/runtime_handoff.json.recovery/experiment_plan.md describing full and proxy options.recovery/source_manifest.json listing only allowed sources.recovery/logs/generated_skill_invocations.json with evidence for every core module.recovery/logs/generated_data_item.json and recovery/logs/training_trace.json for reduced proxy evidence.recovery/recovery_result.json and recovery/experiment_validation.json.Run the bundled tests/test_msd_recovery_protocol.py for source-manifest and mechanism-check helpers. The full recovery harness in the attempt directory should additionally run validate_recovery_experiment.py.
This skill accepts a declared reduced/proxy target only in soft mode. It must never label proxy evidence as full reproduction of the paper figures.
A negative source-boundary check rejected an original_repo source; keep this check as a required recovery-protocol regression.