메뉴
Use when evaluating, scoring, diagnosing, or explicitly refactoring a single skill under controlled governance.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Adapter entry for bootstrapping agent rules, contracts, validation, task packages, reports, and handoff boundaries for financial data engineering or operations projects. Canonical source is in skills/financial-data-agent-bootstrap.
Use when bootstrapping agent rules, contracts, validation, task packages, reports, and handoff boundaries for financial data engineering or operations projects.
Use when onboarding a new maintainer or AI agent, scanning an unfamiliar repository, or generating a takeover packet.
Use when preparing a handoff task package, bounded execution, or execution report across AI agents.
Use when assessing migration readiness for a financial-data Python project with script sprawl, Excel assets, or Wind/Desktop coupling.
Use when updating ai-skill-hub handoff as a system document and the merge must stay section-aware instead of becoming a full rewrite.
| name | skill-governance |
| description | Use when evaluating, scoring, diagnosing, or explicitly refactoring a single skill under controlled governance. |
| metadata | {"triggers":["evaluate a skill against the template","diagnose skill maturity and structure","score a skill and decide rewrite level","govern a single skill before refactor","perform a controlled skill rewrite"],"side_effects":["read_only","write_files"]} |
This execution-focused skill definition keeps the behavior, invocation shape, and adapter-facing contract unchanged while moving explanation-oriented content into supporting assets.
这个 pattern 的核心是：
evaluatedeciderefactor(optional)它稳定组织为：evaluate -> decide -> refactor(optional)。
Input:
rewrite=true/falseProcess:
evaluatedeciderefactor(optional)Output:
skill_assessment_output where applicable: maturity_score, evidence, inference, open_questions, risk_priority, impact_scope, next_action这样组织的原因是 skill 治理首先是控制问题，而不是内容生成问题。先 evaluate 再 decide，可以把修改建立在可复核证据上；把 refactor 设计成可选阶段，则能确保治理动作保持 Non-intrusive，不会默认改变现有 skill。
默认只评估，不修改。
Evaluate first, do not modify by default.
修改必须显式触发。
Rewrite must be explicitly requested.
一次只处理一个 skill。
One skill at a time.
默认不改变项目。
Do not modify the project unless explicitly allowed.
评分要基于结构与证据。
Score against structure and evidence.
重写必须保持非侵入。
Keep refactoring non-intrusive.
evaluate：读取单个目标 skill。
读取目标目录中的 README.md、SKILL.md、相关 prompt 或 agent 文件，确认其是否完整、是否符合 SKILL_TEMPLATE.md 的 10 段结构、以及其主要类型和职责边界。
diagnose：生成治理诊断。
输出 SCORECARD、DIAGNOSIS 和 LEVEL，明确结构缺口、原则偏差、类型错位、prompt 不可执行点和潜在风险，不在此阶段做任何改写。治理 assessment 应按 shared assessment output protocol 在适用时标注 maturity_score、evidence、inference、open_questions、risk_priority、impact_scope 和 next_action。
decide：决定是否进入重写。
根据评分和诊断判断是否需要重写；若 rewrite=false，则停留在 dry-run 治理结果；若 rewrite=true，则明确本次只允许对当前 skill 做受控重写。
refactor(optional)：执行可选受控重写。
仅在显式授权时，基于 SCORECARD 与 DIAGNOSIS 进行最小必要重写，要求对齐 SKILL_TEMPLATE.md、对齐 skill 类型，并保持 Non-intrusive 原则，不扩展到其他 skill 或项目逻辑。
执行链路固定为：evaluate -> diagnose -> decide -> refactor(optional)。
dry-run，未显式授权时只允许评估与诊断rewrite=true 时发生，且应是最小必要改动