Back to Details

cross-evolution

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Cross-Evolution

跨进化

Purpose

目标

Maintain a high evolutionary standard across all agent skills by identifying "Genes" (best practices), transferring them to skills that lack these features, and discovering new high-value genes from existing skills.
通过识别「基因」（最佳实践）、将其转移至缺少这些特性的Skill、并从现有Skill中发现新的高价值基因，来维持所有Agent Skill的高进化标准。

Axioms

基本原则

1. Atomic Independence (No Cross-Skill Glue)

1. 原子独立性（无跨Skill绑定）

Skills and their scripts must be 'atomic'.
  • 'FORBIDDEN': Any logic inside a skill's script that explicitly checks for or calls tools from another skill.
  • 'The Agent is the Glue': Only the AI Agent is responsible for coordination.
Skill及其脚本必须具备「原子性」。
  • 禁止：Skill脚本中任何显式检查或调用其他Skill工具的逻辑。
  • Agent是绑定层：仅AI Agent负责协调工作。

2. Pragmatic Evolution (Occam's Razor)

2. 务实进化（奥卡姆剃刀原则）

  • 'Just-In-Time Transfer': A gene is transferred only when it solves a recurring problem.
  • 'KISS Compliance': Avoid bloating skills with genes they don't need.
  • 'Fitness-Driven': Prioritize genes with highest weight for maximum fitness gain.
  • 即时转移：仅当基因能解决反复出现的问题时，才进行转移。
  • 遵循KISS原则：避免为Skill添加不需要的基因导致臃肿。
  • 适配性驱动：优先选择能带来最高适配性提升的高权重基因。

3. Living Protocol

3. 动态协议

This skill proactively evolves itself and its gene registry upon discovering new constraints.
该Skill会在发现新约束时主动自我进化并更新其基因注册表。

Core Concepts

核心概念

Gene

基因

A modular implementation or documentation pattern that improves skill quality. Genes have a 'lifecycle': 
Proposed → Active → Deprecated → Extinct
. Full registry with weights, detection rules, and conflicts: 
docs/genes.md
一种能提升Skill质量的模块化实现或文档模式。 基因拥有生命周期：
Proposed → Active → Deprecated → Extinct
。 包含权重、检测规则和冲突信息的完整注册表：
docs/genes.md

Horizontal Gene Transfer (HGT)

水平基因转移（HGT）

Copying a gene from a donor skill to a recipient without rewriting the recipient's core purpose.
在不改变接收方Skill核心目标的前提下，将基因从供体Skill复制到接收方Skill。

Gene Discovery

基因发现

Scanning existing skills for repeated high-value patterns not yet represented in the registry. If a candidate passes thresholds, it is automatically written into 
docs/genes.md
under "Proposed Genes".
扫描现有Skill，寻找尚未在注册表中体现的重复出现的高价值模式。 如果候选基因达到阈值，会自动写入
docs/genes.md
的「候选基因」部分。

Genetic Drift

基因漂变

Genes carried by zero skills are candidates for deprecation. After sustained non-adoption, they go extinct.
没有任何Skill使用的基因会被标记为待弃用。长期未被采用的基因会被淘汰。

Fitness

适配性

Quantitative health score per skill: 
earned_weight / applicable_weight × 100%
. Domain-specific genes (recommend=
none
) only count if already present — reward, not penalty.
每个Skill的量化健康评分：
earned_weight / applicable_weight × 100%
。 领域特定基因（推荐值=
none
）仅在已存在时计入评分——这是奖励而非惩罚。

Recombination

基因重组

When two genes conflict in the same skill, selective pressure creates a new hybrid gene.
当同一Skill中的两个基因发生冲突时，选择性压力会催生新的混合基因。

Operating Modes

运行模式

Mode 1: Scan & Audit

模式1：扫描与审计

'Automated': 
bash "${SKILL_DIR}/scripts/audit-genes"
Produces: Gene × Skill matrix, fitness scores, genetic drift warnings, conflict alerts, recommendations, and discovered gene candidates (with registry sync).
Useful flags:
  • --no-discovery
    — skip discovery phase.
  • --no-sync-discovery
    — discover candidates without writing to registry.
自动化执行：
bash "${SKILL_DIR}/scripts/audit-genes"
生成内容：基因×Skill矩阵、适配性评分、基因漂变警告、冲突警报、改进建议，以及发现的候选基因（并同步至注册表）。
实用参数：
  • --no-discovery
    — 跳过基因发现阶段。
  • --no-sync-discovery
    — 发现候选基因但不写入注册表。

Mode 2: Mutation (Update)

模式2：变异（更新）

Inject missing/extra genes into a target skill. Prioritize by weight × fitness impact.
'Value guard': Before transferring a gene, ask: "Does this gene solve a real problem the skill has encountered, or are we just making the fitness number go up?" If the latter — skip the transfer. Fitness score is a heuristic, not a goal. Optimizing the score instead of skill quality is the Farmville trap.
向目标Skill注入缺失或补充基因。按权重×适配性影响优先级排序。
价值校验：在转移基因前，需确认：「该基因是否能解决Skill实际遇到的问题，还是仅仅为了提高适配性分数？」如果是后者，则跳过转移。适配性分数是一种启发式指标，而非目标。为了优化分数而忽视Skill质量属于「Farmville陷阱」。

Mode 3: Speciation (Creation)

模式3：物种形成（创建）

Create a new gene/skill by combining existing patterns (triggered by gene conflicts or recombination).
通过组合现有模式创建新的基因/Skill（由基因冲突或重组触发）。