divergent-thinking-skill 1.0.0

package/README.md

@@ -0,0 +1,49 @@
+# divergent-thinking-skill
+
+npm package that installs the **divergent-thinking** agent skill (`SKILL.md`) so users can run:
+
+```bash
+npx divergent-thinking-skill
+```
+
+## Install
+
+Default target: `~/.agents/skills/divergent-thinking/SKILL.md`.
+
+```bash
+# Custom parent (folder that contains per-skill subdirs)
+npx divergent-thinking-skill --dir ~/.codex/skills
+npx divergent-thinking-skill --dir ~/.claude/skills
+```
+
+Or set environment (parent directory only):
+
+```bash
+DIVERGENT_THINKING_SKILL_DIR=~/.claude/skills npx divergent-thinking-skill
+```
+
+## Publish to npm (maintainers)
+
+1. Edit `package.json` → set `"repository.url"` to your real Git URL.
+2. Pick an available name on npm; if `divergent-thinking-skill` is taken, use a scoped name:
+   `@your-scope/divergent-thinking-skill` and update the `"name"` field.
+3. From this directory:
+
+   ```bash
+   npm login
+   npm publish --access public
+   ```
+
+   `prepublishOnly` runs `scripts/sync-skill.js` and refreshes `skill/SKILL.md` from the repo root `skills/divergent-thinking/SKILL.md`.
+
+## Local test without publishing
+
+```bash
+cd packages/divergent-thinking-skill
+npm run sync
+node bin/install.js --dir /tmp/skills-test
+```
+
+## License
+
+MIT

package/bin/install.js

@@ -0,0 +1,81 @@
+#!/usr/bin/env node
+/**
+ * Installs packaged SKILL.md into the user's agent skills directory.
+ * Usage:
+ *   npx divergent-thinking-skill
+ *   npx divergent-thinking-skill --dir ~/.codex/skills
+ *   DIVERGENT_THINKING_SKILL_DIR=~/.claude/skills npx divergent-thinking-skill
+ */
+const fs = require("fs");
+const path = require("path");
+const os = require("os");
+
+function expandHome(p) {
+  if (!p || p[0] !== "~") return p;
+  return path.join(os.homedir(), p.slice(1).replace(/^\//, "") || "");
+}
+
+function parseArgs(argv) {
+  let outDir = null;
+  let help = false;
+  for (let i = 2; i < argv.length; i++) {
+    const a = argv[i];
+    if (a === "--help" || a === "-h") help = true;
+    else if (a === "--dir" && argv[i + 1]) {
+      outDir = expandHome(argv[++i]);
+    } else if (a.startsWith("--dir=")) {
+      outDir = expandHome(a.slice("--dir=".length));
+    }
+  }
+  return { outDir, help };
+}
+
+function defaultSkillsRoot() {
+  const env =
+    process.env.DIVERGENT_THINKING_SKILL_DIR ||
+    process.env.AGENT_SKILLS_DIR ||
+    "";
+  if (env) return expandHome(env.trim());
+  return path.join(os.homedir(), ".agents", "skills");
+}
+
+function main() {
+  const { outDir, help } = parseArgs(process.argv);
+  if (help) {
+    console.log(`
+divergent-thinking-skill — install SKILL.md for agent runtimes (Cursor, Codex, Claude Code, etc.)
+
+Usage:
+  npx divergent-thinking-skill
+  npx divergent-thinking-skill --dir ~/.agents/skills
+  npx divergent-thinking-skill --dir ./my-project/.cursor/skills
+
+Environment:
+  DIVERGENT_THINKING_SKILL_DIR   Parent directory for skill folders (default: ~/.agents/skills)
+  AGENT_SKILLS_DIR               Same as above (fallback alias)
+
+Install location:
+  <parent>/divergent-thinking/SKILL.md
+
+Point your tool at the parent "skills" directory per its docs (e.g. ~/.codex/skills, ~/.claude/skills).
+`);
+    process.exit(0);
+  }
+
+  const parent = outDir || defaultSkillsRoot();
+  const destDir = path.join(parent, "divergent-thinking");
+  const src = path.join(__dirname, "..", "skill", "SKILL.md");
+  const dest = path.join(destDir, "SKILL.md");
+
+  if (!fs.existsSync(src)) {
+    console.error("install: packaged SKILL.md missing at", src);
+    process.exit(1);
+  }
+
+  fs.mkdirSync(destDir, { recursive: true });
+  fs.copyFileSync(src, dest);
+  console.log("Installed:", dest);
+  console.log("Parent skills directory:", path.resolve(parent));
+}
+
+main();

package/package.json

@@ -0,0 +1,34 @@
+{
+  "name": "divergent-thinking-skill",
+  "version": "1.0.0",
+  "description": "Agent skill: diverge → compare → recommend (frontend-friendly). Install SKILL.md via npx.",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/YOUR_ORG/mult-solutions.git",
+    "directory": "packages/divergent-thinking-skill"
+  },
+  "keywords": [
+    "agent",
+    "skill",
+    "cursor",
+    "claude",
+    "divergent-thinking",
+    "frontend"
+  ],
+  "bin": {
+    "divergent-thinking-skill": "bin/install.js"
+  },
+  "files": [
+    "bin",
+    "skill",
+    "README.md"
+  ],
+  "scripts": {
+    "sync": "node scripts/sync-skill.js",
+    "prepublishOnly": "node scripts/sync-skill.js"
+  },
+  "engines": {
+    "node": ">=18"
+  }
+}

package/skill/SKILL.md

@@ -0,0 +1,104 @@
+---
+name: divergent-thinking
+description: >
+  Use when the user wants multiple approaches, option comparison, feasibility or “best” picks—
+  especially front-end/UI (组件方案、状态管理、样式、性能、无障碍、SSR/SSG、设计还原、工程化).
+  Also general open questions: 多个方案、对比一下、brainstorm、trade-offs、方案对比、可行性.
+  Produces distinct options, a shared-criteria matrix, and a labeled primary recommendation plus fallback.
+---
+
+# Divergent thinking (multi-option → compare → recommend)
+
+## When to use
+
+Apply this skill when:
+
+- The user explicitly asks for several paths, comparison, or what is “most feasible / optimal,” **or**
+- The question is open-ended (how to achieve X) and more than one credible approach exists,
+
+unless the user clearly wants a **single short answer** only.
+
+**Primary domain:** **Front-end engineering** (UI structure, styling systems, client performance, accessibility, framework/library choice, design handoff, delivery trade-offs). For non-UI questions, keep the same workflow with domain-appropriate criteria.
+
+**With `brainstorming` skill:** For greenfield product/feature ideation, you may use the brainstorming skill first to clarify intent and constraints, then use **this** skill to produce comparable options and a recommendation.
+
+**With UI skills:** When output must be visually or structurally strong, combine with **`frontend-design`** / **`ui-ux-pro-max`** (or similar) for layout/tokens before or while building the matrix.
+
+## Workflow (follow in order)
+
+1. **Restate** the goal, success criteria, and hard constraints in your own words.
+2. **Clarify or assume:** If missing facts would change the ranking (budget, timeline, risk tolerance, metrics, platform limits), ask up to **three** targeted questions **or** proceed with a labeled **Assumptions** block.
+3. **Diverge:** Produce **3–5** options for broad/strategic questions, or **2+** for narrow technical questions. Each option must differ in **mechanism or posture** (build vs. buy vs. partner), not wording alone.
+4. **Compare:** Build a **matrix or table**—every option × the **same** criteria. For **front-end** topics, include engineering-relevant columns such as **runtime/bundle impact**, **accessibility**, **maintainability**, **fit to current stack**, **design-handoff friction**, **migration cost**, plus user goals (ship speed, quality). For non-UI topics, use domain-appropriate columns only. Do not compare on different axes per row.
+5. **Recommend:** Give a **primary** pick and name what “best” means here (e.g. *best under stated priorities*, *most feasible now*, *lowest risk*). If criteria conflict, name a **runner-up** or **conditional** choice (“if speed matters more than cost, pick B”).
+6. **Uncertainty:** If facts are thin, add **Unknowns** and **What would change the ranking** (specific evidence or experiments).
+
+**Optional — tools first:** If the host has MCP or web access, prefer **fetching or searching** for missing facts *before* step 3, or mid-workflow when a claim is checkable. If tools are unavailable, keep **Unknowns** explicit. (See also **Optional integrations** below.)
+
+## Quality gates (before you send)
+
+- **Orthogonality check:** Can each option be summarized in one line without overlapping the others? If two collapse into the same approach, replace one.
+- **No fake diversity:** Do not list the same idea with different adjectives.
+- **Explicit trade-offs:** Where speed vs. cost vs. quality conflict, say which criterion drove the primary recommendation.
+- **High-stakes caution:** For medical, legal, financial, or safety-critical decisions, add a **Risk note**: not a substitute for a qualified professional; do not imply a single “optimal” legal/medical/financial outcome without verification.
+
+## Output skeleton
+
+Use this structure (headings can be localized to the user’s language):
+
+1. **Summary** — One short paragraph: problem, number of options, top pick + sense of “best.”
+2. **Problem & constraints** — Restatement + **Assumptions** (if any).
+3. **Options** — For each: name, one-line approach, expected outcome, main cost/risk.
+4. **Comparison** — Table/matrix: rows = options, columns = shared criteria (use ✓/– or L/M/H if helpful).
+5. **Recommendation** — Primary + **labeled “best” type**; runner-up or conditional branch; **Unknowns** + **what would change the ranking** when relevant.
+6. **Risk note** (if high-stakes or irreversible) — Scope of caution and suggestion to verify with a human expert.
+
+## Installation
+
+- **Repo-local (this project):** `skills/divergent-thinking/SKILL.md` — reference from project `AGENTS.md` or Cursor rules if you want it always considered.
+- **User-wide:** Copy the folder to `~/.agents/skills/divergent-thinking/` (or your tool’s skills directory) so the agent loads it globally.
+
+## Optional integrations (public MCPs & skills)
+
+This skill does not require any server. To improve **grounded** comparisons, combine with tools your environment already exposes:
+
+| Need | Role | Where to browse |
+|------|------|-----------------|
+| Grounding / fetch / search | Turn “feasible” into “checked against a source” | [MCP Registry](https://modelcontextprotocol.io/registry), [registry.modelcontextprotocol.io](https://registry.modelcontextprotocol.io/), [modelcontextprotocol/servers](https://github.com/modelcontextprotocol/servers), [MCP Directory](https://mcpserverdirectory.org/) |
+| Browser automation | Validate flows that depend on live UI | Registry categories (browser / testing) |
+| Git / GitHub | Scope and risk for **this** repo | Git/GitHub-class MCPs in the reference repo above |
+| Memory / notes | Persist user constraints across sessions | Registry “knowledge / memory” style servers — mind privacy |
+
+**Skill ecosystem (no single store):** [Agent Skills spec](https://agentskills.io), example gallery [anthropics/skills](https://github.com/anthropics/skills). Use **brainstorming**-style skills *before* this one for discovery; use domain skills *after* shortlisting.
+
+### Front-end–oriented MCP patterns (when available)
+
+| Need | Role | Typical MCP patterns |
+|------|------|----------------------|
+| Design source | Ground options in real components/layouts | **Figma**-class MCP (design → structure / tokens) |
+| Browser truth | Layout, perf, network, a11y checks | **Chrome DevTools**-class MCP |
+| E2E / UI flows | Scripted navigation, screenshots, regressions | **Playwright**-class MCP |
+| Docs | API and version truth for frameworks | **Fetch** + official docs |
+| Repo | Match existing stack and conventions | **Git** / **GitHub** MCP |
+
+Use **stack → (optional) design/repo/docs via MCP → diverge → matrix → recommend.** If tools are missing, state **Unknowns**.
+
+Full rationale: `openspec/changes/divergent-thinking-skill/proposal.md` (*Complementary public skills & MCPs*, *Front-end scenarios & complementary tools*).
+
+## Verification walkthrough (spec coverage)
+
+Use these as internal self-checks; outputs should satisfy the scenarios in `openspec/changes/divergent-thinking-skill/specs/divergent-thinking-skill/spec.md`.
+
+| Type | Example prompt | You must |
+|------|----------------|----------|
+| Broad / strategic | “我们团队效率低，怎么系统性提升？” | ≥3 distinct mechanisms, 3–5 options, full matrix, recommendation with “best” type |
+| Narrow technical | “Python 里用哪个 HTTP 客户端更合适？” | ≥2 real alternatives or justify single viable path; same comparison columns for each |
+| Conflicting goals | “要快上线还要零技术债，怎么选架构？” | Call out trade-off; primary + fallback when priorities differ; optional Unknowns |
+| Front-end open | “中后台表格页，数据量大，怎么设计前端架构？” | ≥3 distinct approaches; matrix includes perf, a11y, maintainability, stack fit |
+| Front-end narrow | “Tailwind vs CSS Modules vs styled-components，选哪个？” | ≥2 options; include bundle/DX/a11y/migration criteria |
+| Front-end + design | “Figma 还原 + 主题切换，给几种实现路线？” | Mention handoff/tokens; if Figma MCP available, read structure before options |
+
+## Trigger phrases (non-exhaustive)
+
+中文：多个方案、几种思路、对比一下、哪个更可行、优缺点、权衡、可行性、最优方案、发散；前端常用：组件方案、状态管理、样式方案、性能、无障碍、SSR、设计稿还原、工程化选型。  
+English: brainstorm options, compare approaches, trade-offs, pros and cons, most feasible, best path, alternatives; UI: styling strategy, state management, SSR vs SPA, a11y, bundle/perf, component architecture.