菜单
Reference for creating and editing skills. Covers format, naming, frontmatter, progressive disclosure, idle registration, and dead zone handling. Read before creating any skill.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
| name | skill-creation |
| description | Reference for creating and editing skills. Covers format, naming, frontmatter, progressive disclosure, idle registration, and dead zone handling. Read before creating any skill. |
| allowed-tools | Read, Write, Edit, Glob, Grep |
Read this before creating or editing any skill. For the full template, see TEMPLATE.md.
local/skills/<name>/SKILL.md — Your skills. Always create here..claude/skills/<name>/SKILL.md — Framework skills. Never modify.---
name: my-skill # lowercase, hyphens only, matches directory name
description: Does X when Y. # Third person. What it does + when to use it.
allowed-tools: Read, Edit # Tools this skill needs (comma-separated)
model: sonnet # Optional: haiku, sonnet, opus
---
## When to Use — Required for chat skills. Without it, the skill is idle-only.## Idle Behavior + Cooldown: N minutes — Required for idle skills. Without both, the skill won't auto-run.Keep SKILL.md under 100 lines. Move detailed content to reference files in the skill directory:
my-skill/
SKILL.md # Overview + navigation (loaded when skill triggers)
REFERENCE.md # Detailed docs (loaded only when needed)
FORMAT.md # Data format examples
SKILL.md references them: "See REFERENCE.md for details." Claude reads reference files only when needed — zero token cost until accessed. Keep references one level deep (no nested references).
If a skill detects nothing to do until a future time, edit its own Cooldown: line to a longer value. The skill loader hot-reloads every idle cycle. Edit back to normal when the dead zone passes.
Idle runs have no conversation output. Use send_to_channel with channel_id: "group".
Invoke this skill before creating a new skill or making significant edits to an existing one.
Tags transcript message threads with semantic concepts (roast, game-recommendation, etc.) so past conversations can be searched by idea, not just keywords.
Captures notable interactions missed during live conversation — new person info, game preferences, boundaries, jokes that landed/flopped, relationship insights.
The single owner of the hypothesis lifecycle — creates, tests, promotes, and kills hypotheses. Uses extensions as programmatic data collectors and reviews their output.
Consolidates impression files by applying weight decay, removing redundant entries, and enforcing per-user limits. Maintains relationship file summaries.
Consolidates old memories by removing duplicates, promoting recurring patterns to learned-patterns.md or relationship files, and archiving outdated entries.
Identifies behavioral patterns from recent conversations and reaction feedback, updates learned-patterns.md when 3+ observations confirm a pattern.