Verified:GitHub

xiushen-lu

Analyze runtime metric logs and select evolution plans with adaptive thresholds.

10|1|Updated May 4, 2026

Search Tags:#bottleneck detection #self-evolution #runtime metrics #rollback safety #adaptive thresholds #cross-skill learning #deterministic tuning

Ready to run

npx skills add https://github.com/isLinXu/under-one --skill xiushen-lu

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: xiushen-lu
Source: https://github.com/isLinXu/under-one/tree/main/underone/skills/xiushen-lu
Command: npx skills add https://github.com/isLinXu/under-one --skill xiushen-lu

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

This Skill solves the problem of agent skill degradation over time by analyzing runtime performance metrics and proposing controlled self-improvements without breaking existing behavior.

Core Features & Use Cases

Adaptive threshold engine: Detects stability, error-proneness, volatility, and autonomy pressure, then dynamically adjusts decision thresholds with boundary “护栏”.
Deep evolution with rollback safety: Performs parameter-level tuning/extension/refactor across the target skill’s scripts, creates backups, validates outcomes, and rolls back automatically on failure.
Cross-skill learning & threshold migration: Shares validated threshold-related learnings via a knowledge hub so improvements can propagate safely across skills.

Quick Start

Run the core evolution engine against your skills directory to generate an evolution report and update adaptive thresholds only when explicitly applied.

Frequently Asked Questions about xiushen-lu

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I stop LLM agent skills from degrading over time in production?▼

To stop LLM agent skills from degrading over time, you need runtime metric analysis to detect drift and apply controlled self-improvements. This Skill evaluates execution logs to propose parameter-level tuning and refactoring while maintaining existing behavior.

How does adaptive threshold tuning work for agent skill maintenance?▼

Adaptive threshold tuning works by detecting stability, error-proneness, volatility, and autonomy pressure from runtime metrics, then dynamically adjusting decision boundaries with guardrails. This ensures parameter-level tuning only triggers when validated conditions are met.

What is the best way to safely rollback script changes during skill evolution?▼

Safe rollback during skill evolution requires creating backups before applying parameter-level tuning or refactoring to scripts, validating outcomes post-change, and automatically reverting to the backup if validation fails. This prevents broken behavior from persisting.

Can I share validated threshold tuning improvements across multiple agent skills?▼

Yes, you can share validated threshold tuning improvements across multiple agent skills using a cross-skill knowledge hub. This allows threshold migration so learned stability improvements propagate safely to other skills experiencing similar runtime drift.

When do I need deterministic tuning for production-style skill maintenance?▼

You need deterministic tuning for production-style skill maintenance when repeated skill execution causes quality degradation, increased errors, or higher human intervention rates. It applies a validated core engine to ensure code evolution remains backed and reversible.

How do I detect bottlenecks causing error-proneness in my agent skill scripts?▼

To detect bottlenecks causing error-proneness in agent skill scripts, the system classifies runtime metric logs against adaptive thresholds to identify volatility and autonomy pressure. This bottleneck classification directs targeted parameter-level tuning and refactoring.