What problem does it solve? Understanding which knobs Verl exposes to users requires manually reading dozens of example shell scripts and YAML files. This Skill automates that discovery by scanning Verl examples for environment variables, Hydra overrides, and CLI flags, then maintaining a Chinese-language parameter intelligence library with per-parameter explanations and an Excel ledger. ## Core Features & Use Cases - Automated Parameter Scanning: Parses examples/**/*.sh and examples/**/*.yaml for FOO=${FOO:-default} environment overrides, Hydra keys like actor_rollout_ref.*, and --flag arguments, classifying each as 配置, 算法, or 效率. - Per-Parameter Explanation Files: Generates and reuses markdown files under docs/verl/features/lists/ with Chinese explanation, common values, performance impact, accuracy impact, and evidence snippets. - Subagent Research Batches: Splits unresolved 待补充 parameters into batch files so subagents can research official Verl docs and repo docs, then re-runs the scan to absorb results. - Use Case: Point the scan script at a local Verl checkout to produce docs/verl/features/verl-example-parameters.xlsx listing every example parameter with category, Chinese explanation, common values, performance and accuracy impact, and example count. ## Quick Start Ask the agent to scan the local Verl repository examples and generate the Chinese parameter explanation library plus the Excel ledger under docs/verl/features.
