Research Experimenter

技能档案

Research Experimenter

Autonomous, fixed-budget optimization of target logic. Use when you need to run iterative experiments (mutations) against a target file to improve its performance or reliability within a strict time limit (The Karpathy Standard).

Clinscott0 星标2026年3月24日

职业
分类: 框架内部

技能内容

🔱 Research Experimenter Skill [Ω]

This skill implements the "Karpathy Standard" for autonomous optimization through bounded experimentation.

Core Mandates

The Crucible: Every experiment is bound by a fixed time budget (default: 300s).
Fixed Evaluation: The metric and test environment are immutable constants.
Hypothesis First: You MUST document a hypothesis before making a code change.
Neuralplastic Evolution: Successes are only accepted if they show a statistically significant improvement (use sprt-verifier).

Workflow

Hypothesize: Identify a specific improvement (e.g., "Add skip connections to improve gradient flow").
Mutate: Modify the target file (the "mutable genome").
Train/Test: Run the evaluation loop for the fixed budget duration.
: Analyze the resulting metric (e.g., validation loss, bits-per-byte).