philosophy-refinement
Use only when the user explicitly asks to change, review, or persist reusable development philosophy.
jyt6640 · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add jyt6640/persona-harness --skill philosophy-refinement技能说明
name: philosophy-refinement description: Use only when the user explicitly asks to change, review, or persist reusable development philosophy. persona-skill: core mutability: explicit-user-action handoff: null
Socratic Philosophy Refinement
Start only when the user explicitly asks to change, review, or persist a reusable philosophy. Ordinary chat, code inspection, a direct code-change request, design criticism, and implicit preference signals do not start this skill.
Read the relevant approved decisions and code first. Ask only for missing material rationale, alternatives, scope, counterexamples, or trade-offs, one question at a time. Reassess what remains unresolved after each answer; there is no mandatory order or question count. Explain the current question when the user is unsure, honor stop immediately, and reuse already approved facts and scope without asking again. Do not infer missing rationale or exceptions.
A topic represents one decision dimension. Complementary obligations may use distinct precise topics; alternative choices for the same dimension stay on the same topic. Different wording is not proof of contradiction. The V1 resolver treats overlapping same-topic records conservatively as unresolved, not as a semantic judgment. Explain the concrete overlap and preserve all required meaning in any user-approved consolidation. Never invent topic names merely to bypass a real conflict or silently supersede a rule.
Classify a complete explicit outcome as an implementation mistake, a
project/task decision, or a personal philosophy candidate. Only a complete
candidate may be handed to ph philosophy refine --stdin as the structured
personalization-refinement.v1 input. The CLI reuses the append-only profile
store: a non-conflicting candidate may activate, while an overlap remains
pending for explicit retain, exception, supersede, or pending resolution.
Read the persistence contract only at that approved
write boundary; it contains the existing structured formats and plugin bridge.
Do not write files, profile state, workflow state, plans, issues, branches, or agents during the conversation. Never persist raw prompts, responses, code, credentials, or absolute paths. This procedure is host-neutral and does not require OpenCode or a model action.