193 hqhq1025
skill-personalizer Skill
在审计或改造新建、下载、fork、安装或社区的 Agent Skills,使其适配用户的工具、习惯、目录、会话历史和偏好工作流时使用,尤其是触发条件感觉不对、太嘈杂或太宽泛时。
安装方式:把技能目录放入 ~/.claude/skills/(Claude Code)或在 claude.ai 设置中启用;也可复制右侧安装命令一键添加。
技能指令原文(SKILL.md)
Skill Personalizer
Overview
Audit and tune a skill for one user's real environment. The goal is not public portability; it is better triggering, less friction, and stronger fit with the user's actual tools, paths, style, and recurring tasks.
When To Use
- A user installs a skill from GitHub or creates one from scratch and wants it to fit their setup.
- A skill undertriggers, overtriggers, asks unnecessary questions, or misses the user's preferred workflow.
- A user asks whether an existing skill is good, broken, noisy, too long, conflicting, or worth keeping.
- Local paths, aliases, memories, CLIs, MCP tools, or repo conventions should be reflected in the skill.
Do not use when preparing a skill for public release; use skill-generalizer for that.
Workflow
- Inspect the target skill, installed copies, local memories, and real session evidence when available.
- Run the audit checks in audit-rubric.md when quality, trigger fit, or retention is unclear.
- Identify the user's recurring phrasing, expected autonomy level, tools, directories, and verification habits.
- Compare the skill's trigger conditions against real user requests that should or should not load it.
- Edit only the target skill and bundled resources needed for personalization.
- Add concrete local defaults, preferred commands, safety boundaries, and verification steps.
- Preserve useful upstream behavior; document any intentional local divergence.
- Validate with realistic prompts and a frontmatter/layout check.
Personalization Rules
- Personal details are allowed only if they improve this user's future execution.
- Do not add brittle fallbacks that hide broken local setup.
- Prefer real local evidence over generic best practices.
- Keep trigger descriptions broad enough to catch the user's natural phrasing.
- If editing an installed third-party skill, avoid changing upstream attribution or license text.
References
Read audit-rubric.md for the diagnostic pass inherited from the original optimizer.
Read personalization-rubric.md for local defaults, session evidence, and validation scenarios.