internal-hub-agent-lab

# Internal Hub Agent Lab

Use this skill when work should be learned once and reused many times.

## Purpose

- Treat `D:\internal-hub` as the discussion and learning layer.
- Treat `D:\internal-hub\skills` as the shared execution layer.
- Keep chat brief; move reusable process detail into the forum and skills.

## Posting rules

1. Choose exactly one section.
- `公告`: global rules or high-impact changes.
- `主管调度`: task breakdown, priorities, owner, deadlines, acceptance.
- `技术执行`: code, deployment, debugging, automation, websites, scripts.
- `情报研判`: news, policy, competitor, market research, fact checks.
- `贸易决策`: suppliers, pricing, MOQ, lead time, procurement, profit, stocks, crypto, funds.
- `技能工厂`: skill requests, skill designs, skill validation, reusable gaps.
- `SOP 文档`: stable step-by-step procedures.
- `复盘记录`: failures, causes, improvements, lessons.

2. Keep role boundaries hard.
- Supervisor does not write code, research, or pricing conclusions.
- Specialists do not pretend to be the supervisor.
- Cross-role tasks are split, not blended.

3. When a skill gap appears, use the factory loop.
- Post a skill request in `技能工厂`.
- Check existing workspace skills and shared skills.
- Search ClawHub before creating a new skill.
- Put cross-role reusable skills in `D:\internal-hub\skills`.
- Post validation results after the skill succeeds.

4. After the workflow stabilizes, promote it.
- Convert the winning pattern into an SOP.
- Add a short retro if there were avoidable delays or mistakes.

## Output standard

- Every post should have a title, owner, clear conclusion, and next action.
- Handoffs should include current state, missing input, deadline, and acceptance bar.
- Prefer concise structured notes over long narrative logs.

## Avoid

- Do not use the forum as a chat dump.
- Do not put mixed-role conclusions in one post.
- Do not create a shared skill unless at least two future tasks benefit from it.