social-research
Skill1 starsChanged 25 days agoApache-2.0View on GitHub
从一个研究方向和一份或多份数据,产出一篇社科定量论文。串起数据画像、文献检索、找研究空白、可行性检验、计量分析、论文写作全流程。用于「做个社科研究」「帮我从这份数据出一篇论文」「社科选题到成稿」等请求。
6 other projects use a version of this.
Adopted by
Most recently taken up by
- ali-demirbas/researchtheir own version, first seen Nov 16, 2026
- robertmnyborg/everyday-agent-skillstheir own version, first seen Nov 15, 2026
- openSocialScience/openSocialScienceword for word, first seen Oct 25, 2026
- tylerprogramming/claude-skillstheir own version, first seen Oct 15, 2026
- trac3r00/agent-skillstheir own version, first seen Aug 26, 2026
- smcllns/skillstheir own version, first seen Jun 16, 2026
Other versions6 projects use a version of this.
Word for word · 1
- openSocialScience/openSocialScience★ 1100% same
Edited · 0
None found.
Wrote their own version · 5
- ali-demirbas/research★ 0same name
- robertmnyborg/everyday-agent-skills★ 0same name
- smcllns/skills★ 0same name
- trac3r00/agent-skills★ 0same name
- tylerprogramming/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- social-ideaSkill
- social-identificationSkill
- social-paperSkill
- social-reviewSkill
- social-simSkill
The file
Files
社科定量研究主链 · 路由器
原始输入:$ARGUMENTS
如果上面显示的是字面的 $ARGUMENTS,说明宿主不替换占位符;原始输入就是用户消息里 /social-research 之后的全部内容。
**这份文件只负责路由。**每个环节的契约在 stages/,跨环节的硬规矩在 rules/,散文、教训、例子在
stages/<环节>.notes.md。用 Read 工具按相对本文件的路径读,不要凭记忆做事。
第一步:拆参数
**用户不必学任何符号。**三种进来的方式都要认——什么都没打(直接进访谈,一样一样问)、
一句人话(自己理解,再复述一行等他点头)、带 --- 的完整写法(机械拆)。
访谈怎么问、为什么「你问的是关于国家的事还是关于人的事」比问 data: 有用:
读 rules/input-and-rundir.md。
带 --- 时这么拆:研究方向 --- data: <文件,逗号分隔,或 pack:<名>> --- discipline: <学科> [--- lang: en|zh] [--- style: <刊名>] [--- refs: <自带文献:题录文件 / PDF / PDF 目录,逗号分隔>] [--- idea: <你已经有的想法原文>] [--- design: <五个键>] [--- type: 因果|描述|测量|仿真] [--- checkpoint: idea,ideas,design,none]
- data 没给就停下来问用户,别猜;
discipline缺省economics;lang缺省zh——产出中文论文 + 中文期刊体例的图(贯穿环节 5、7:稳健性 label 用中文、图--lang zh、/social-paper带--- lang: zh)。用户说要写英文论文才传en,那时图也跟着走英文体例。文献检索不受它影响(中英文献都要) - 带
idea:就是自带想法模式:环节 3 不出题,只把用户那条想法整理成唯一的idea_001;design:给的键锁进plan.json。拆法、design:五个键与plan.json字段的映射表:读 rules/input-and-rundir.md - 多份数据怎么写、
pack:怎么拿、对照表怎么传:读 rules/input-and-rundir.md - 一律复述一行给用户确认:
方向「XXX」|数据 x.csv + y.csv|学科 economics|语言 zh(语言 en 可省)。 说人话进来的必须等他点头再往下——理解可能读错,读错了他一眼看得见, 不复述就是跑了七十分钟才发现用错了数据
第二步:定运行目录
**每个 oss 命令都要带同一个 --run;永远不要 cd。**理由与两次事故见 rules/input-and-rundir.md。
mkdir -p runs/r20260829_1530 # 名字你定,之后不再改第三步:读状态,能续就续
开工先看运行目录里有没有 run_state.json:
cat <运行目录>/run_state.json 2>/dev/null; oss loop status --run <运行目录>- 有,且
stage_status不是done→ 从stage指的那个环节续跑,不从头来 - 有
anchor→ 那就是本次要回答的问题,不许改;换题只能oss anchor switch --to <idea_id> --why ... --authorized-by user review.round是真实轮数,以它为准,不凭记忆- 没有 → 新开
stage是空的(环节 1–5 不写状态)→ 看目录里已有哪些产出(profile.json→literature.json→screening.json→reading_cards.json→ideas.json→power.json→plan.json→results.json),从最后一份的下一环节续,别重跑已有的- 这一步只看
阶段/锚点/轮次三行;决定那一行只在环节 8 之后才作数
上下文被压缩之后也一样:先读这个文件,再决定下一步。
第四步:按 manifest 加载
读 manifest.yaml,照它做三件事:
always_load里的五份rules/每次都读——它们是跨环节的硬规矩- 按
run_state.stage在axes.stage.values里查到本环节的契约,只读这一份 - 契约的「完成条件」全满足才推进到下一环节;卡住了再按
references.on_demand读对应的.notes.md
| stage | 环节 | 契约 |
1-profile | 数据画像 | stages/1-profile.md |
2-literature | 文献检索与精读 | stages/2-literature.md |
3-ideas | 找空白、想 idea、补数据 | stages/3-ideas.md |
4-feasibility | 绑构念、算功效 | stages/4-feasibility.md |
5-plan | 写分析计划 | stages/5-plan.md |
6-run | 跑分析与处置 | stages/6-run.md |
7-paper | 写论文 | stages/7-paper.md |
7.5-identification | 因果设计审计(研究类型是因果、从因果降级来的,或仿真货箱时;细则见该环节「输入」) | stages/7.5-identification.md |
8-review | 审稿—改—复审 | stages/8-review.md |
9-wrapup | 自评与收尾 | stages/9-wrapup.md |
跨环节的硬规矩(always_load)
| 文件 | 管什么 |
| rules/iron-laws.md | 四条铁律:判断归你计算归 oss、不写死宿主、你只写四个文件、正文不许裸数字 |
| rules/input-and-rundir.md | 参数怎么拆、数据怎么拿、运行目录的两条脾气 |
| rules/stop-conditions.md | 循环以 oss loop status 第一行的决定为准;穷尽之后保留设计 |
| rules/ask-user.md | 提问白名单;三个 checkpoint 开关 |
| rules/explore.md | 想试一试只走 oss run --explore --for;不许用 python 直接跑回归 |
开工前先跑一次
oss doctor**这个系统还在分阶段建设中,部分环节返回假数据。**它会告诉你哪些环节已经是真的、哪些还是假的。
不要写死在记忆里——每完成一个阶段这个名单就变。产出里带 FAKE_ 前缀的地方就是还没换真的。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

