anxi-skill 是做什么的？
教练，我想训练 Skill。A local Agent Skill that trains unstable skill packages with bounded patches, human review, validation logs, and rejected memory.
Overview
This skill helps you train and stabilize unreliable skill packages through a controlled cycle of patching, review, logging, and rejection tracking. It is intended for use with local agents (e.g., Claude Code) to manage skill development iteratively.
Features
- Bounded patches: Apply limited, targeted fixes to unstable skills.
- Human review: Get manual approval before applying significant changes.
- Validation logs: Automatically record results of each patch or training step.
- Rejected memory: Store and recall failed attempts to avoid repeating mistakes.
Usage
- Load the skill into your agent.
- Identify an unstable skill package.
- Run the training cycle: patch, validate, review, and iterate.
- Check validation logs and rejected memory for insights.
Compatibility
- Works with Claude Code and similar local agent environments.
- Requires access to local filesystem for skill packages and logs.
Example
# Hypothetical command to train a skill package
anxi train my-unsupported-skill --patch --review --log
Notes
- Patches are bounded to prevent unintended side effects.
- Human review is mandatory for critical updates.
- Rejected memory helps the agent learn from past failures.