技能 Tuning

数据社区

作者仓库星标 2,029
作者仓库 Claude-Code-Workflow

解读按原文结构重写，命令、链接、术语均保留；右侧可核对作者原始 SKILL.md

方法与流程

Workflow：Step · Action · Orchestrator Decision · Output 1 · action-init · status='pending' · Backup, session created 2 · action-analyze-requirements · After init · Required dimensions + coverage

适用与边界

当前原文没有单列适用场景。
当前原文没有单列不适用场景或限制。

原文中的明确线索

要点：「P0」、「P1」、「P2」、「P3」、「P4」、「P5」、「Setup」、「Analysis」
文件与命令：action-init、action-analyze-requirements、action-gemini-analysis、action-generate-report、action-propose-fixes、action-apply-fix、action-complete、.workflow/.scratchpad/skill-tuning-{ts}/state.json

流狐整理：以上内容来自当前 SKILL.md 的章节与原词；未补写作者没有声明的工具、兼容性或能力。

Skill Tuning

Autonomous diagnosis and optimization for skill execution issues.

Architecture

┌─────────────────────────────────────────────────────┐
│  Phase 0: Read Specs (mandatory)                    │
│  → problem-taxonomy.md, tuning-strategies.md         │
└─────────────────────────────────────────────────────┘
                        ↓
┌─────────────────────────────────────────────────────┐
│  Orchestrator (state-driven)                         │
│  Read state → Select action → Execute → Update → ✓ │
└─────────────────────────────────────────────────────┘
        ↓                           ↓
┌──────────────────────┐   ┌──────────────────┐
│  Diagnosis Phase     │   │ Gemini CLI       │
│  • Context          │   │ Deep analysis    │
│  • Memory           │   │ (on-demand)      │
│  • DataFlow         │   │                  │
│  • Agent            │   │ Complex issues   │
│  • Docs             │   │ Architecture     │
│  • Token Usage      │   │ Performance      │
└──────────────────────┘   └──────────────────┘
                ↓
        ┌───────────────────┐
        │  Fix & Verify     │
        │  Apply → Re-test  │
        └───────────────────┘

Core Issues Detected

Priority	Problem	Root Cause	Fix Strategy
P0	Authoring Violation	Intermediate files, state bloat, file relay	eliminate_intermediate, minimize_state
P1	Data Flow Disruption	Scattered state, inconsistent formats	state_centralization, schema_enforcement
P2	Agent Coordination	Fragile chains, no error handling	error_wrapping, result_validation
P3	Context Explosion	Unbounded history, full content passing	sliding_window, path_reference
P4	Long-tail Forgetting	Early constraint loss	constraint_injection, checkpoint_restore
P5	Token Consumption	Verbose prompts, state bloat	prompt_compression, lazy_loading

Problem Categories (Detailed Specs)

See specs/problem-taxonomy.md for:

Detection patterns (regex/checks)
Severity calculations
Impact assessments

Tuning Strategies (Detailed Specs)

See specs/tuning-strategies.md for:

10+ strategies per category
Implementation patterns
Verification methods

Workflow

Step	Action	Orchestrator Decision	Output
1	`action-init`	status='pending'	Backup, session created
2	`action-analyze-requirements`	After init	Required dimensions + coverage
3	Diagnosis (6 types)	Focus areas	state.diagnosis.{type}
4	`action-gemini-analysis`	Critical issues OR user request	Deep findings
5	`action-generate-report`	All diagnosis complete	state.final_report
6	`action-propose-fixes`	Issues found	state.proposed_fixes[]
7	`action-apply-fix`	Pending fixes	Applied + verified
8	`action-complete`	Quality gates pass	session.status='completed'

Action Reference

Category	Actions	Purpose
Setup	action-init	Initialize backup, session state
Analysis	action-analyze-requirements	Decompose user request via Gemini CLI
Diagnosis	action-diagnose-{context,memory,dataflow,agent,docs,token_consumption}	Detect category-specific issues
Deep Analysis	action-gemini-analysis	Gemini CLI: complex/critical issues
Reporting	action-generate-report	Consolidate findings → final_report
Fixing	action-propose-fixes, action-apply-fix	Generate + apply fixes
Verify	action-verify	Re-run diagnosis, check gates
Exit	action-complete, action-abort	Finalize or rollback

Full action details: phases/actions/

State Management

Single source of truth: .workflow/.scratchpad/skill-tuning-{ts}/state.json

{
  "status": "pending|running|completed|failed",
  "target_skill": { "name": "...", "path": "..." },
  "diagnosis": {
    "context": {...},
    "memory": {...},
    "dataflow": {...},
    "agent": {...},
    "docs": {...},
    "token_consumption": {...}
  },
  "issues": [{"id":"...", "severity":"...", "category":"...", "strategy":"..."}],
  "proposed_fixes": [...],
  "applied_fixes": [...],
  "quality_gate": "pass|fail",
  "final_report": "..."
}

See phases/state-schema.md for complete schema.

Orchestrator Logic

See phases/orchestrator.md for:

Decision logic (termination checks → action selection)
State transitions
Error recovery

Key Principles

Problem-First: Diagnosis before any fix
Data-Driven: Record traces, token counts, snapshots
Iterative: Multiple rounds until quality gates pass
Reversible: All changes with backup checkpoints
Non-Invasive: Minimal changes, maximum clarity

Usage Examples

# Basic skill diagnosis
/skill-tuning "Fix memory leaks in my skill"

# Deep analysis with Gemini
/skill-tuning "Architecture issues in async workflow"

# Focus on specific areas
/skill-tuning "Optimize token consumption and fix agent coordination"

# Custom issue
/skill-tuning "My skill produces inconsistent outputs"

Output

After completion, review:

.workflow/.scratchpad/skill-tuning-{ts}/state.json - Full state with final_report
state.final_report - Markdown summary (in state.json)
state.applied_fixes - List of applied fixes with verification results

Reference Documents

Document	Purpose
specs/problem-taxonomy.md	Classification + detection patterns
specs/tuning-strategies.md	Fix implementation guide
specs/dimension-mapping.md	Dimension ↔ Spec mapping
specs/quality-gates.md	Quality verification criteria
phases/orchestrator.md	Workflow orchestration
phases/state-schema.md	State structure definition
phases/actions/	Individual action implementations

流狐档案作者与许可取自来源；运行、权限和网络为流狐检测或估算

流狐分类: 数据
作者声明 Agent: 未找到明确声明；不据此推断已兼容或已测试
静态检查: 88 / 100 · 启发式扫描，不代表运行安全
作者 / 版本 / 许可: @catlog22 · 未声明 license
流狐 Token 估算: 低消耗
流狐接入估算: 需简单配置
是否需要外部 API Key: 未发现要求
检测到的系统要求: 未声明
底层运行要求: 未声明
检测到的文件与系统行为: 只读
允许写入 / 修改
Shell 执行
检测到的网络行为: 仅限本地
安装命令数: 无（仅作为资料）

档案由构建时根据 SKILL.md 与安装命令自动衍生，可能与作者实际意图存在差异。

需要注意： 未限定 allowed-tools，默认拥有全部工具权限。

输出预览 skill-tuning.preview

# Usage Examples

# Basic skill diagnosis
/skill-tuning "Fix memory leaks in my skill"

# Deep analysis with Gemini
/skill-tuning "Architecture issues in async workflow"

# Focus on specific areas
/skill-tuning "Optimize token consumption and fix agent coordination"

# Custom issue
/skill-tuning "My skill produces inconsistent outputs"

SKILL.md SKILL.md

# Skill Tuning

Autonomous diagnosis and optimization for skill execution issues.

## Architecture

```
┌─────────────────────────────────────────────────────┐
│  Phase 0: Read Specs (mandatory)                    │
│  → problem-taxonomy.md, tuning-strategies.md         │
└─────────────────────────────────────────────────────┘
                        ↓
┌─────────────────────────────────────────────────────┐
│  Orchestrator (state-driven)                         │
│  Read state → Select action → Execute → Update → ✓ │
└─────────────────────────────────────────────────────┘
        ↓                           ↓
┌──────────────────────┐   ┌──────────────────┐
│  Diagnosis Phase     │   │ Gemini CLI       │
│  • Context          │   │ Deep analysis    │
│  • Memory           │   │ (on-demand)      │
│  • DataFlow         │   │                  │
│  • Agent            │   │ Complex issues   │
│  • Docs             │   │ Architecture     │
│  • Token Usage      │   │ Performance      │
└──────────────────────┘   └──────────────────┘
                ↓
        ┌───────────────────┐
        │  Fix & Verify     │
        │  Apply → Re-test  │
        └───────────────────┘
```

## Core Issues Detected

| Priority | Problem | Root Cause | Fix Strategy |
|----------|---------|-----------|--------------|
| **P0** | Authoring Violation | Intermediate files, state bloat, file relay | eliminate_intermediate, minimize_state |
| **P1** | Data Flow Disruption | Scattered state, inconsistent formats | state_centralization, schema_enforcement |
| **P2** | Agent Coordination | Fragile chains, no error handling | error_wrapping, result_validation |
| **P3** | Context Explosion | Unbounded history, full content passing | sliding_window, path_reference |

…

证据边界与执行链路

作者原文负责流程事实；流狐只索引当前章节、要点、文件与命令。

章节    -> Architecture → Core Issues Detected → Problem Categories (Detailed Specs) → Tuning Strategies (Detailed Specs) → Workflow → Action Reference
要点    -> P0 · P1 · P2 · P3 · P4 · P5 · Setup · Analysis
文件/命令 -> action-init · action-analyze-requirements · action-gemini-analysis · action-generate-report · action-propose-fixes · action-apply-fix · action-complete · .workflow/.scratchpad/skill-tuning-{ts}/state.json
内容 SHA-256 -> cf9a26970891