← All workflow paths
TASK PACKAGE / OPEN SKILLS

WeChat content

Build your WeChat content skill package: choose steps for your available materials, inspect inputs and handoffs, and export a plan for your own AI agent.

TASK → SKILLS → HANDOFF

Start with your task.

Checking AI availability

Describe the result you need. Choose skills, inspect the handoffs, then take the task package to your own agent.

Confirmed materials you already have

Checked research, positioning and briefs will be skipped. Uncheck them to repeat those steps.

AI selection sends your task description, material selections and answers to DeepSeek. Keep source documents in your own agent.

Or choose a path yourself

A plan you can take with you.

  1. The skills your task needs
  2. Materials, steps and handoffs
  3. A complete Markdown task package

Your own agent executes the work. Review each deliverable before moving on.

MODEL RUN / 2026-09-05 / 02 REPEATS

WeChat: source adaptation → article editing

Review: changes required before use

Actual DeepSeek API runs on synthetic Chinese materials. The harness read the skill instructions and references and passed drafts between steps. No browsing or installed Codex, Claude Code or Cursor host was tested. Findings below were reviewed by Codex, not signed off by a marketing owner.

MODEL
deepseek-v4-flash
SKILL VERSION
2.0.0
TOTAL TIME / RUN
14s / 11s
  • Both outputs retained the planned date, unknown price, fact ledger and pending review without quantified performance claims.
  • Final bodies contained about 494 / 286 Chinese characters, below the requested 600–900. Run 1 incorrectly self-reported about 700.
  • Editing stayed close to the upstream draft. This does not establish a benefit over one model call; length and wording still need review.
Inspect the fixed input (Chinese)

合成测试材料,非真实客户。品牌:栖云协作;产品:交接板。已确认事实:共享任务清单、材料链接、负责人、截止日期和 CSV 导出。没有 CRM 集成,没有自动发布。读者:中国 B2B 软件公司的市场负责人。语气务实、清楚,不夸大。本次只将以下素材适配并编辑为 600-900 字公众号稿,含标题候选、摘要、正文、事实对照和待审核事项。素材:活动准备常分散在不同文档中,成员需要反复确认谁负责、用哪版材料、何时截止。交接板将这些事项放在共享清单中;可从清单导出 CSV 供团队核对。2026-10-15 计划开放公开测试,价格未定。不能把一个可能的协作便利变成已验证的效率收益,不得增加客户案例或数字。CTA 是预约演示,链接待补。所有原始材料都在此,不需要联网查证;未知信息写待确认。品牌审核和对外发布由真实负责人完成;本次测试只产出草稿,后续步骤把草稿视为待审核交接材料。

Inspect an output excerpt (Chinese; unedited)
# 公众号稿(草稿,待审核)

## 标题候选

1. 活动准备总在文档里“迷路”?把责任、版本和截止日放进一张共享清单
2. 每次活动都要反复确认“谁负责、用哪版、何时交”?交接板把答案放在一处
3. 活动筹备的混乱,往往不是执行力问题,而是信息散落的问题

## 摘要

活动准备的材料常分散在不同文档中,成员反复确认谁负责、用哪版、何时截止。交接板将这些事项集中到共享任务清单,并支持 CSV 导出供团队核对。2026 年 10 月 15 日计划开放公开测试,价格待定。

---

## 正文

### 活动筹备的混乱,往往不是执行力问题

筹备一场活动,最常见的情况不是没人干活,而是信息散落在不同地方。

方案在 A 文档,物料清单在 B 表格,负责人和截止日期在聊天记录里。成员想确认自己该做什么,得先找到对的文件,再找到对的版本,再找到对的负责人——每一步都在消耗时间。

更麻烦的是,这些信息是流动的。方案改了一版,表格更新了一行,聊天记录里多了一句“这个我来跟”。如果没有一个统一的地方承接这些变化,团队成员就只能靠反复询问来对齐。

### 把分散的信息放进一张共享清单

交接板做的事情很直接:把活动准备中的关键事项放进一张共享任务清单。

每项任务对应明确的负责人和截止日期,相关材料链接也放在同一处。成员打开清单,就能看到“谁负责、用哪版材料、何时截止”——不需要在多个文档之间来回跳转。

清单还支持 CSV 导出。团队可以把数据带出来,按自己的习惯做核对或存档,不改变原有的工作流程。

### 现阶段说明

交接板目前尚未开放公开测试,计划于 2026 年 10 月 15 日开放,价格未定。

需要说明的是:将分散信息集中到共享清单,解决的是信息对齐的便利性问题。至于它能在多大程度上提升团队效率,目前还没有验证数据,我们不做夸大表述。

### 下一步

如果你正在负责市场活动的筹备工作,并且对“把任务、负责人、截止日期放进一张共享清单”这个思路感兴趣,可以预约演示,实际看看交接板如何运作。

---

## 事实对照表

| 正文表述 | 来源素材 | 类型 |
|---|---|---|
| 活动准备材料常分散在不同文档中 | 素材原文 | [F] 已验证事实 |
| 成员需反复确认谁负责、用哪版材料、何时截止 | 素材原文 | [F] 已验证事实 |
| 交接板将事项放在共享任务清单中 | 素材原文 | [F] 已验证事实 |
| 清单含负责人、截止日期、材料链接 | 已确认事实 | [F] 已验证事实 |
| 支持 CSV 导出供团队核对 | 已确认事实 | [F] 已验证事实 |
| 2026-10-15 计划开放公开测试 | 素材原文 | [F] 已验证事实 |
| 价格未定 | 素材原文 | [F] 已验证事实 |
| “解决的是便利性问题,效率提升无验证数据” | 素材原文“不能把可能的协作便利变成已验证的效率收益” | [I] 编辑推断,基于素材限制条件 |
| “方案在 A 文档,物料清单在 B 表格” | 素材原文“分散在不同文档中”的具体化 | [I] 编辑推断,举例说明 |
| 无 CRM 集成、无自动发布 | 已确认事实 | 正文未提及,避免引入无关信息 |

## 待审核事项

1. **标题选择**:三个候选标题需品牌负责人选定或修改。
2. **“效率提升无验证数据”表述**:正文中“至于它能在多大程度上提升团队效率,目前还没有验证数据”为编辑推断,建议保留以体现务实语气,需确认。
3. **示例具体

Excerpt only; the download contains every full response, including incomplete outputs.

Download both runs, input, outputs and hashes ↓

Follow-up: narrower step outputs

Two further runs per task used the same materials with explicit per-step scope and a 6,500-token cap. All responses ended normally. This was a changed harness configuration, not a new skill version or proof of improvement across tasks.

Final bodies contained about 457 / 537 Chinese characters, still below the requested 600–900. A larger output limit did not fix length compliance.

Download both follow-up runs ↓

These runs show where review is needed. They do not prove customer results, traffic gains, superiority to a general agent, or reliability for every plan using these skills.

Before you start

Prepare the materials

  • Interviews, comments, sales notes, or other customer language
  • One source asset with confirmed facts and boundaries
  • Target channels, audience, and publishing objective

Keep human decisions explicit

  • Match the strength of each insight to the source sample
  • Have a local content owner review facts, voice, and publishing risk

Review the actual outputs

Check the facts and handoffs after each step. Retrospectives start only after real results are available. Installation compatibility is not a claim of model performance.

View the method examples →