Skill Comply

affaan-m/ECC/docs/zh-CN/skills/skill-comply

by affaan-mef648e01899ba3e8dc6371642deaaf64b4477775No license275K starsListed Oct 9, 2026Updated Oct 9, 2026Repository updated 4 days ago

可视化技能、规则和代理定义是否被实际遵循——自动生成3种提示严格级别的场景,运行代理,分类行为序列,并报告完整工具调用时间线的合规率

Instructions onlyAI & Agents
AI-generated overview

Measures whether coding agents actually follow skills, rules, or agent definitions, and reports compliance rates.

What it does
It generates an expected behavior specification from a given .md file, produces scenarios at three decreasing prompt-strictness levels (supportive, neutral, competitive), runs the agent and captures tool-call traces, classifies tool calls into specification steps with an LLM, checks ordering deterministically, and outputs a self-contained report with compliance scores and timelines.
When to use it
Use it after adding a new rule or skill to verify agent compliance, when someone asks whether a rule is really followed, or as periodic quality maintenance.
Requirements
Requires the uv Python runner, the claude CLI for running prompts, and model access for generation and classification; the skill is instructions only and ships no scripts, though it references a scripts.run module.

skill-comply:自动化合规性测量

通过以下方式测量编码代理是否实际遵循技能、规则或代理定义:

  1. 从任意 .md 文件自动生成预期行为序列(规范)
  2. 自动生成提示严格程度递减的场景(支持性 → 中性 → 竞争性)
  3. 运行 claude -p 并通过 stream-json 捕获工具调用轨迹
  4. 使用 LLM(而非正则表达式)将工具调用分类到规范步骤
  5. 确定性检查时间顺序
  6. 生成包含规范、提示和时间线的自包含报告

支持的目标

  • 技能(skills/*/SKILL.md):工作流技能,如搜索优先、TDD 指南
  • 规则(rules/common/*.md):强制性规则,如 testing.md、security.md、git-workflow.md
  • 代理定义(agents/*.md):代理是否在预期时被调用(内部工作流验证尚不支持)

何时激活

  • 用户运行 /skill-comply <path>
  • 用户询问"这条规则是否真的被遵循?"
  • 添加新规则/技能后,验证代理合规性
  • 作为质量维护的一部分定期执行

使用方法

bash
# Full runuv run python -m scripts.run ~/.claude/rules/common/testing.md
# Dry run (no cost, spec + scenarios only)uv run python -m scripts.run --dry-run ~/.claude/skills/search-first/SKILL.md
# Custom modelsuv run python -m scripts.run --gen-model haiku --model sonnet <path>

关键概念:提示独立性

测量技能/规则是否在提示未明确支持时仍被遵循。

报告内容

报告是自包含的,包括:

  1. 预期行为序列(自动生成的规范)
  2. 场景提示(每个严格程度级别询问的内容)
  3. 每个场景的合规性评分
  4. 带有 LLM 分类标签的工具调用时间线

高级(可选)

对于熟悉钩子的用户,报告还包含针对合规性较低的步骤的钩子提升建议。此为参考信息——主要价值在于合规性本身的可见性。

Source and attribution

Source:affaan-m/ECCindocs/zh-CN/skills/skill-complyat commitef648e0

License: No license

Content belongs to its original authors. SourceWeft indexes it from a public repository.

Report or request removal