菜单
Preload repo context for the dancing-bear personal-assistants codebase. Use when the user runs /familiarize, says "get familiar with this repo", or starts a new session and wants Claude oriented before making changes.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/dancing-bear-show/dancing-bear --skill familiarize命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Adversarial swarm PR code review. Fetches diff, sweeps for relevant concerns, fans out per-file validators, consolidates findings, and posts inline GitHub comments.
Run adversarial swarm review then auto-fix all findings in parallel. Use when the user says "review and fix", "swarm fix", or wants review and fix in one shot.
Fetch every PR review thread (Copilot and human), triage each one, fix them in parallel, verify with happy/sad-path coverage, refresh the PR title and description, then reply to and resolve every thread. Use when the user says "address the review threads", "fix the review comments", "handle the Copilot feedback", or "resolve the threads on PR N".
基于 SOC 职业分类
正在显示 SKILL.md
| name | familiarize |
| description | Preload repo context for the dancing-bear personal-assistants codebase. Use when the user runs /familiarize, says "get familiar with this repo", or starts a new session and wants Claude oriented before making changes. |
| allowed-tools | Bash, Read, Glob |
| skills | ["dancing-bear-rules"] |
Load context about dancing-bear at session start.
./bin/llm familiar --stdout
This returns a YAML capsule covering:
Load these only when the task requires deeper understanding:
| File | When to Load |
|---|---|
.llm/CONTEXT.md | Architecture questions, system overview |
.llm/MIGRATION_STATE.md | In-flight work, current status |
.llm/PATTERNS.md | Implementing features, copy-paste templates |
.llm/FAMILIARIZE_CORE.md | Familiarization contract details |
Generate on demand (never preload):
./bin/llm domain-map --stdout # directory structure deep dive
For CLI capabilities, use agentic schemas (never --help):
./bin/<assistant> --agentic --agentic-format yaml --agentic-compact
Primary surfaces: mail-assistant, calendar-assistant, schedule-assistant, phone-assistant.
Load only the wrapper(s) relevant to the current task.
Visualization + orchestration surfaces (load when the task involves rendering or DAG orchestration):
./bin/charts — render time-series charts from JSON (render, grid, reshape)./bin/diagrams — Mermaid diagrams (from-yaml, render, validate, embed, telemetry cost/token pies)./bin/workflow — YAML DAG workflow engine (parse, compile, run, lint, list, status)These three are standalone argparse CLIs, not llm --app routes, so they have no agentic schema — discover them with <wrapper> --help (the one exception to the "never --help" rule above; llm familiar --verbose surfaces them the same way).
You should know:
./bin/<assistant> <subcommand> <flags>)charts, diagrams, workflow).llm/PATTERNS.md).venv/, .git/, .cache/, maker/, _disasm/, out/, _out/, backups/)/familiarize, /dancing-bear-rules, etc.)