Skills for Real Engineers. Straight from my .agents directory.
science-campaign-design
Agent skill fromyzlin286-wq/Deepscient
Use when a research goal is too large for one agent turn and must be split into schedulable work — when planning a campaign across S0–S11, when deciding how fine a shard should be (literature, statistics, figures, writing), when setting gates and vote counts so parallel results can be trusted, or when a batch plan produced duplicates, idle workers or unusable outputs.
How to install science-campaign-design
npx skills add yzlin286-wq/Deepscient# or add only this skill to Claude Code: copy the folder
dsh-science/skills/science/science-campaign-design/
# into
~/.claude/skills/science-campaign-design/ (all projects)
.claude/skills/science-campaign-design/ (one project)
1# Campaign 设计:从研究问题到可分片的任务23## 四层拆解45```6Campaign(一个课题的一次冲顶)7 └─ Workstream(S0–S11 中的一段,有独立 gate)8 └─ Task(一个可验收的交付,有 owner 与 gate)9 └─ Shard(一次模型调用能做完的最小单元)10```1112- **Campaign** 与课题一一对应,有唯一 `campaignId`;跨课题不复用。13- **Workstream** 必须有 `stage`(S0…S11)与 `gate`;没有 gate 的 Workstream 不允许存在。14- **Task** 是派单单位,必须有唯一 owner(L2 岗位)与 `acceptance`。15- **Shard** 是执行单位,必须有稳定 `idempotencyKey`,同一 key 重跑不得产生第二份产物。1617## 拆到哪一层为止1819判断标准只有一条:**这一层能否单独验收**。2021- 能验收 → Task。22- 不能验收,但能并行 → 继续拆到 Shard。23- 拆完还说不清"什么叫做完" → 不是粒度问题,是目标没定义,回 S0。2425## 分片粒度经验值2627| 分片类型 | 一片的量 | 理由 |28| --- | --- | --- |29| 文献 | 50–200 条记录/片 | 低于 50 上下文浪费;高于 200 单次摘要质量下降 |30| 全文精读 | 3–8 篇/片 | 需要逐段引用时按篇切,不要按段落切 |31| 统计 | 1 个检验/片 | 假设检查与效应量必须同片,拆开就对不上 |32| 作图 | 1 张图/片 | 尺寸、字号、误差棒说明是整图属性 |33| 写作 | 1 节/片 | 引言/方法/结果/讨论各自风格不同,混写必串味 |34| 写作-润色 | 800–1500 字/片 | 超过就要拆,否则一致性无法检查 |35| 代码执行 | 1 个命令/片 | 便于定位失败与重跑 |3637**不许按"平均分"切**。1000 条文献分 5 片是 200 条/片,分 20 片是 50 条/片,都合法;38分 50 片是 20 条/片,属于过度分片,协调成本超过收益。3940## Shard 的最小字段4142```json43{44 "shardId": "lit-0031",45 "taskId": "task-lit-map",46 "role": "shard-worker",47 "kind": "literature",48 "input": { "query": "...", "window": "2019-2026", "offset": 150, "limit": 200 },49 "acceptance": { "minRecords": 150, "requireDoi": true },50 "idempotencyKey": "task-lit-map:2019-2026:150:200",51 "maxAttempts": 3,52 "votes": 153}54```5556`idempotencyKey` 由 `taskId + 分片输入指纹` 组成,不含时间戳、不含随机数。57包含时间戳的 key 等于没有幂等。5859## Gate 怎么设6061一个 Task 的 gate 必须满足三条:62631. **可机器判**:来自工具输出,不是"读起来还行"。例如 `science_stats` 返回了64…
Excerpt shown under the repository's MIT license.
More skills in Deepscient
- ▸science-agent-scheduling
- ▸science-consensus-report
- ▸science-discipline-methods
- ▸science-experiment-runbook
- ▸science-figure-for-journal
- ▸science-frontier-to-paper
- ▸science-fulltext-hunting
- ▸science-journal-intel
- ▸science-journal-selection
- ▸science-literature-search
- ▸science-manuscript-formatting
- ▸science-org-roles
Similar skill repositories
An agentic skills framework & software development methodology that works.
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
Public repository for Agent Skills
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
An AI skill that provides design intelligence for building professional UI/UX across multiple platforms.
