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 agentic-feature-design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 | Agentic Feature Design |
| description | Designing features for the "Action Era" that are AI-accessible by default |
Features in LivestockAI must now be designed for dual consumption: Humans (UI) and Agents (MCP/API).
Every feature must be fully functional via Server Functions before any UI is built.
Test: Can an agent complete the entire user story using only bun run check-feature-x.ts (a script calling server functions)?
If yes -> Agent Ready.
If no (logic lives in React components) -> Refactor immediately.
Server functions are the tools we give to agents. Document them accordingly.
export const createBatchFn = createServerFn({ method: 'POST' })
.inputValidator(batchSchema)
.handler(async ({ data }) => {
/* ... */
})
/**
* @description Creates a new livestock batch.
* @workflow
* 1. Check `getFarmFacilities` for space.
* 2. `createBatchFn`
* 3. Log initial `feed_record` if applicable.
*/
Agents operate on Intent, not just data.
Instead of generic CRUD (updateBatch), expose semantic actions (graduateBatch, quarantineBatch).
// ❌ Generic
updateBatch(id, { status: 'sold' })
// ✅ Semantic (Agent Friendly)
markBatchAsSold(id, { date, price, customer })
Semantic actions allow agents to:
Agents need a standard way to ask for permission for high-stakes actions (e.g., ordering feed, selling stock).
The ApprovalRequest Entity:
agentId: Who is asking?action: JSON payload of the server function to call.summary: Human-readable explanation ("I want to order 50 bags of Starter Feed").status: PENDING | APPROVED | REJECTEDAll major entities (Batches, Sales) should have an ai_metadata JSONB column.
Agents use this to store reasoning ("Predicted harvest date change due to low feed intake").
Do not show this raw JSON to users, but use it to power UI "Insights".
three-layer-architecture - Where the logic livesfeature-structure - How to organize the files