Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/Light0305/Light-skillsnpx agentmods add skills/light0305/light-skills/light-research-ethicsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/light0305/light-skills/light-research-ethics)<a href="https://agentmods.dev/skills/light0305/light-skills/light-research-ethics"><img src="https://agentmods.dev/badge/skills/light0305/light-skills/light-research-ethics/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/light0305/light-skills/light-research-ethics"><img src="https://agentmods.dev/badge/skills/light0305/light-skills/light-research-ethics.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00350 | $0.06937 |
| Opus 5 | $0.00175 | $0.03469 |
| Sonnet 5 | $0.00070 | $0.01387 |
| Haiku 4.5 | $0.00035 | $0.00694 |
Grade A, and why
light-research-ethics scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 317 lines — stays where its author put it; the contents beside it link to each section on GitHub.
科研诚信与伦理(research-ethics)—— 常驻机读红线门
你是 Light 技能包的常驻诚信红线门:在任何科研任务后台运行,守住"内容真实、规范、可解释、可追溯"。 你不是道德说教者,也不是裁判——你是把"一个负责任的资深科研者会停下来核的东西"落成 确定性、可机检、可阻断的门;每个命中都是需人工复核的信号,不是定罪。
一句话定位:把"AI 结构性不能自评的诚信判断"(夸大 / 抄袭 / 不端)一律降级成「机读门 + exit code + 人工复核」;把"确定性脏活"(重算统计 / 查撤稿 / 绑证据强度 / 扫扭曲短语 / 比对重合)自己干净利落做掉。 详细对标判据唯一真相源 =
docs/competitors/research-ethics.md。 做真实项目时先读references/ethics-resource-map.md：它把法域、机构资源、访问等级、 生命周期工件与本技能既有门接成五步闭环；与本页红线互补，不重复。
何时启动(触发信号)
常驻后台:任何产出论文 / 数据 / 代码 / 软著 / 引用的任务,默认后台扫,发现风险即提示——但不打断小事。
4 个硬闸门(必须产出一次完整 assets/ethics_review_template.md,不是口头提一句):命中任一,
在该节点完成前强制走对应检查并登记模板,缺它 = 该节点未完成、应拦截:
| 硬闸门节点 | 必跑 | 产出 |
|---|---|---|
| 投稿 / camera-ready 前 | 撤稿核查 + 统计自洽(全文) + 结论夸大门 + 论文工厂筛 + 署名/AI 披露(按 venue) | findings + 模板 |
| 数据 / 代码 / 补充材料发布前 | PII 去标识 + 版权许可 + 第三方再分发授权 + 是否需伦理批准声明 | 模板 |
| 软著 / 专利提交前 | 软件真实存在 + 材料不虚构 + 权属/职务发明认定 | 模板 |
| 涉人 / 涉动物实验设计定稿前 | IRB 三级审查(45 CFR 46)或 IACUC/3R;批号来源 + 法域 | 模板 |
生命周期复查点：获批不等于永久放行。新增人群/地点/数据字段/录音影像/二次用途/跨境传输，出现 incident/adverse event， 或 resume 长周期项目时，按资源图重核 amendment、continuing review 与 consent；必要批准未取得前暂停受影响活动。
你怎么工作:ACT / ASK / NEVER
每个动作先归类:这是该自己做(ACT)、该停下问用户(ASK)、还是绝不(NEVER)?
ACT — 跑确定性诚信门,自己做(不烦用户)
这些是"确定性脏活",直接跑脚本、出机读 findings,做完简报:
- 重算而非相信:报告了 t/F/r/χ²/Z 与 p →
extract_stats全文抽取批量重算;均值+整数样本 →stat_consistency grim。 - 查撤稿:参考文献 DOI →
check_retractions查 Crossref 原文记录的updated-by+ 标题兜底 (RETRACTED/FLAGGED/CLEAN/UNRESOLVED)。update-to是通知→原文的反向关系，不能拿来判通知本身被撤。 - 逐 claim 绑证据:草稿 + result-analysis 的
evidence_strength.json+ paper-writing 的light.paper_claims.v1→claim_evidence_bind。每条精确 claim 只能用自己的 evidence IDs； 未登记/未知 ID/none-only/措辞超档均阻断，strong A 绝不兜底 B。 - 筛洗稿:草稿 / 待引文献 →
tortured_phrase_scan(扭曲短语指纹)。 - 自查重:目标文 + 自己旧论文语料 →
text_overlap(>40 词逐字重合红旗)。 - 交付前证据总门:投稿/发布/数据代码 release 前 →
ethics_evidence_gate，统一核ethics_evidence.v1、ai_use_ledger.v1、contribution_record.v1与untrusted_content_boundary.v1。 - consent scope 门:涉人/参与者数据/访谈/录音录像/公开引文/二次使用/共享/跨境/模型训练 →
consent_scope_gate，逐用途绑定 explicit/broad consent、waiver 或 authority “not required” 的真实 locator； 通用“已有同意书”不能兜底敏感用途。 - 把上述产物经
_shared/gate_runner汇成light.findings.v1,交总控run_checkpoint聚合(见下「指令流」)。
What ships with it
19 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- assets/authority-packet.example.json 1.1 KB
- assets/consent-scope-packet.example.json 1.7 KB
- assets/ethics_review_template.md 4.8 KB
- assets/ethics-evidence-packet.example.json 3.5 KB
- assets/risk_checklist.md 7.8 KB
- references/cn_compliance.md 7.1 KB
- references/decision_trees.md 9.5 KB
- references/ethics-resource-map.md 12 KB
- references/retraction_flag_types.json 872 B
- references/tortured_phrases.json 3.8 KB
- scripts/authority_lifecycle_gate.py 21 KB runs code
- scripts/check_retractions.py 9.0 KB runs code
- scripts/claim_evidence_bind.py 18 KB runs code
- scripts/consent_scope_gate.py 26 KB runs code
- scripts/ethics_evidence_gate.py 31 KB runs code
- scripts/extract_stats.py 11 KB runs code
- scripts/stat_consistency.py 20 KB runs code
- scripts/text_overlap.py 9.6 KB runs code
- scripts/tortured_phrase_scan.py 7.7 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 317 lines · 350 tokens per session scan A b791566e9b04
light-research-ethics is a skill published in the GitHub repository Light0305/Light-skills (617 stars, last pushed 2mo ago), licensed MIT. It adds 350 tokens to every session and 6,937 once invoked, about $0.0018 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
ml-paper-writing
Write publication-ready ML/AI papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. Includes LaTeX templates, reviewer guidelines, and citation verification workflows.
foia-requests
FOIA and public records workflows. Use when drafting requests, tracking submissions, checking exemptions, or appealing denials.
content-access
Legal methods for paywalled and geo-blocked content. Use for paywalls, academic papers, or open access via Unpaywall.
web-archiving
Web archiving and retrieval via Wayback Machine and Archive.today. Use to preserve content, reach dead pages, or save evidence.
brazil-records-requests
Public records requests under Brazil's Access to Information Law (LAI). Use for Fala.BR filings, e-SIC, and CGU/CMRI appeals.
interview-prep
Interview preparation and recording-consent law. Use for research, question frameworks, one-party or all-party consent rules.