majiayu000
majiayu000/claude-skill-registry
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/majiayu000/claude-skill-registry --skill brainstorm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。
LLM token logprobs and calibration. Per-decision confidence, ECE, Brier, reliability diagrams, low-confidence triage.
Analyze LLM token logprobs and calibration. Use for per-decision confidence, ECE, Brier scores, reliability diagrams, and low-confidence triage.
回顾最近 N 天的 Claude Code 使用记录——扫描原始会话数据，按主题分组汇总"我都做了什么"，并从个人操作系统视角输出模式、风险与增删建议。当用户说 /recap、"看看我这几天做了什么"、"回顾一下我最近的会话"、"这两天我用 claude 干了啥"、"活动回顾" 时使用。
基于 SOC 职业分类
正在显示 SKILL.md
| name | brainstorm |
| description | Separate goals from implementation. |
| practices | ["lean-startup","mythical-man-month"] |
| hexagonal_role | domain |
| consumes | ["standards"] |
| produces | ["result.json","verdict.json"] |
| context_rel | [{"kind":"shared-kernel","with":"standards"}] |
| skill_api_version | 1 |
| metadata | {"tier":"execution","dependencies":[]} |
| context | {"window":"inherit","intent":{"mode":"none"},"sections":{"exclude":["INTEL","HISTORY","TASK"]},"intel_scope":"none"} |
| output_contract | skills/research/schemas/findings.json |
Purpose: Separate WHAT from HOW. Explore the problem space before committing to a solution.
Upstream of move 1 (shape intent as BDD) of the operating loop. Consumes a free-text goal; produces Given/When/Then-shaped acceptance examples that /discovery can fold into a BDD intent issue. The Capture step (phase 4 below) is not complete until at least one happy path and one critical edge are written as testable Gherkin — "it should work" is not a captured example.
Four phases:
/plan/brainstorm "add user authentication" # full 4-phase process
/brainstorm # prompts for goal
If the user provided a goal string, evaluate it. Otherwise prompt for one.
Use AskUserQuestion with options to gauge clarity:
If vague or exploring, ask follow-up questions to sharpen the goal before proceeding. Do NOT move to Phase 2 until you have a concrete problem statement (one sentence, testable).
Answer these questions (use codebase exploration as needed):
Summarize findings before moving on. If anything is unclear, ask the user.
Generate 2-3 distinct approaches. For each:
Before asking the user to choose, stress-test each approach:
For each approach, answer these red team questions (read references/red-team-checklist.md):
Mark any approach that fails 2+ red team questions as HIGH RISK in the comparison.
If all approaches fail 2+ questions, generate a 4th "hybrid" approach addressing the weaknesses.
Present the comparison and use AskUserQuestion to let the user pick an approach or request a hybrid.
Generate a date slug: YYYY-MM-DD-<goal-slug> (lowercase, hyphens, no spaces).
Write the output file to .agents/brainstorm/YYYY-MM-DD-<slug>.md:
---
id: brainstorm-YYYY-MM-DD-<goal-slug>
type: brainstorm
date: YYYY-MM-DD
---
# Brainstorm: <Goal>
## Problem Statement
## Approaches Considered
## Selected Approach
## Open Questions
## Next Step: /plan
All five sections must be populated. The "Next Step" section should contain a concrete /plan invocation suggestion with the selected approach as context.
Create the .agents/brainstorm/ directory if it does not exist.
Phase 4 output written = done. No further phases, no loops.
After writing the output file, verify:
Problem Statement, Approaches Considered, Selected Approach, Open Questions, Next Step: /plan) are present and non-emptyReport the file path to the user.
Example 1: Specific goal
User: /brainstorm "add rate limiting to the API"
Phase 1: Goal is clear — add rate limiting to the API.
Phase 2: Problem is uncontrolled request volume causing timeouts.
Benefits operators and end users. No rate limiting exists today.
Phase 3: Three approaches — token bucket middleware, API gateway,
per-route decorators. User picks token bucket.
Phase 4: Writes .agents/brainstorm/2026-02-17-rate-limiting.md
Example 2: Vague goal
User: /brainstorm "improve performance"
Phase 1: Goal is vague. Asks: "Which part? API response times,
build speed, database queries, or something else?"
User says: "API response times on the search endpoint."
Phase 2: Investigates search endpoint, finds N+1 queries.
Phase 3: Approaches — query optimization, caching layer, pagination.
Phase 4: Writes .agents/brainstorm/2026-02-17-search-performance.md
| Problem | Cause | Solution |
|---|---|---|
| Brainstorm loops in Phase 1 without advancing | Goal remains too vague after follow-up questions | Provide a concrete, testable problem statement (e.g., "reduce API search latency below 200ms" instead of "improve performance"). |
| Output file missing one or more required sections | Phase 4 was interrupted or the skill terminated early | Re-run /brainstorm with the same goal; verify all 5 sections (Problem Statement, Approaches Considered, Selected Approach, Open Questions, Next Step: /plan) are present in the output. |
.agents/brainstorm/ directory not created | The skill could not create the directory (permissions or path issue) | Manually create it with mkdir -p .agents/brainstorm and re-run. |
/plan invocation in "Next Step" section is generic or incomplete | The selected approach was not specific enough to generate a concrete plan command | Edit the output file to refine the selected approach, then craft a /plan invocation that includes the approach name and key constraints. |
| Brainstorm produces only one approach in Phase 3 | The problem space is narrow or the goal is overly constrained | Widen the goal slightly or explicitly ask for alternative approaches (e.g., "consider a caching approach and a query optimization approach"). |