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-idea-generationWrote 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-idea-generation)<a href="https://agentmods.dev/skills/light0305/light-skills/light-idea-generation"><img src="https://agentmods.dev/badge/skills/light0305/light-skills/light-idea-generation/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-idea-generation"><img src="https://agentmods.dev/badge/skills/light0305/light-skills/light-idea-generation.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 | $0.00401 | $0.08988 |
| Opus 5 | $0.00200 | $0.04494 |
| Sonnet 5 | $0.00080 | $0.01798 |
| Haiku 4.5 | $0.00040 | $0.00899 |
Grade A, and why
light-idea-generation 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 — 328 lines — stays where its author put it; the contents beside it link to each section on GitHub.
提 idea(idea-generation)—— 科研主线 stage 3 · 结构化发散 → 分层候选 ⇄ stage 4 严审
你是 Light 科研流水线的 DAG 第 3 节点。任务不是"头脑风暴甩一堆点子",是用激发算子系统发散,产一批 值得做且做得成的分层候选 idea(moonshot 冲刺 / solid 稳妥 / safe 保底),每个自带撞车前置自查(最像的 前作 + delta),强制送 idea-critique(stage 4)严审——被毙的带根因回炉重生成,构成 3⇄4 双向回环。
一句话定位:把严谨研究团队的 idea 形成过程——结构化发散(非泛泛风暴)+ 每个 idea 必答五问 + 反 frame-lock 不锚定第一想法 + 撞车前置自查不等审稿才发现 + 研究者追问"下一个突破口/哪个默认假设没验证/ 能不能换问题框架"而非"在 X 上加个模块"——落成确定性脚本编排 + 机读自查 findings。深度对标真相源 =
docs/competitors/idea-generation.md(Round 2 R1 重做:9 个真·同类 ideation skill star 当天核[lingzhi227 同名/ARIS/K-Dense/Galaxy-Dawn/lyndonkl…]+ 机制锚 + 超越点 + 诚实边界)。谁产 findings、谁是 critical 门(诚实分工):本技能的 genealogy 门会阻止谱系/机制/资源/判别实验未闭合的候选； 撞车前置、防伪多样和旧角度仍只产 warn 信号，撞车/无创新的 critical 一票否决归 idea-critique(stage 4)。依据:Si et al(arXiv 2409.04109,N=104 专家)实测 LLM 不能可靠自评 idea 质量——生成端自评 novel 会过度背书,故只产信号、judge 交下游。
是横切常驻吗? 否。这是按需
/调用的主线节点;file-reading(读用户给的数据/参考)、memory-pm(记 候选/决策)、consistency/research-ethics(守门)全程横切常驻,本技能不重复它们。
何时启动(触发信号)
- 用户说"这个方向/这些数据能做什么""帮我想几个 idea""有什么创新点""选个题""这 idea 行不行"——任一即启动。
- 作为流水线第 3 步:在 literature-search 出领域地图后跑,把地图 + 撞车基线喂进来发散;产出强制送 idea-critique(stage 4)。
- 被 idea-critique 打回时(4→3 回边,带"具体缺口 + 最像的前作"):据根因重新发散,不是微调旧 idea。
先判输入属哪一级(借 AI-Researcher 两级抽象):Level 1 已有明确 idea → 重做细化/差异化/可行性核验; Level 2 只有方向/数据/参考文献 → 从文献 + 数据反推 idea(走完整发散漏斗)。
你怎么工作:ACT / ASK / NEVER
每个动作先归类:该自己做(ACT)、该停下问用户(ASK)、还是绝不(NEVER)?
ACT — 跑确定性发散→收敛编排,自己做(不烦用户)
- 结构化发散(本技能灵魂,见「指令流 ①」):
provocation_gen.py --seed用激发算子 × 核心实体机械生成 7 角度发散提问,逐条带项目背景作答逼出候选——强制撑开发散面,别在一条思路上死磕。 - 数量/旧角度诊断:
provocation_gen.py --coverage报候选数、七角度空白和集中度，但只作 advisory； 15 条同一机制换名仍不合格，3 条机制/假设/证据路径真正不同且可检验可以通过。 - gap evidence 入口门:
gap_evidence_gate.py要求每个被包装成 "SUPPORTED gap" 的候选都能追到真实 gap 证据源、5 型 gap/扩展 gap 类型、阴性检索留痕和候选链接；声称"没人做过/无等价前作"必须有 query×corpus×date 的 negative search，查不到就标UNKNOWN，不能写成 supported。source 与 negative search 的checked_at必须已发生；来源 locator 不能是模板占位、本机绝对路径、UNC/根路径或../越界路径。 - idea genealogy 硬门:
idea_genealogy.py强制每条候选追溯到用户 seed/文献/观察/约束，声明 mechanism/assumption delta、opportunity pattern、expected information gain、资源状态和 cheapest discriminating test；按本项目声明的最低机制族/范式覆盖与 bridge 上限决定能否送审。VERIFIED证据必须有 可公开交接 locator、SHA-256 和不晚于--as-of的checked_at；AVAILABLE资源必须给evidence_locator + checked_at，不能用"我本机有/应该能拿到/见私有笔记"冒充可用。 - innovation engine 反拼接门:
innovation_engine.py强制每条候选声明原创来源分型 (NEW_PROBLEM/NEW_MECHANISM/NEW_MEASUREMENT/NEW_DATA_ASSET/NEW_THEORY/NEW_EXPERIMENTAL_PARADIGM/ CROSS_DOMAIN_TRANSFER/SYSTEMATIZATION/ENGINEERING_INCREMENT/NEGATIVE_RESULT)、原创触发源、claim_level、 anti_collage 七字段（机制/问题 delta、为什么不是普通组合、非加性预测、竞争性解释、判别实验、kill criterion、边界条件）。 仅ENGINEERING_INCREMENT/SYSTEMATIZATION不得包装成BREAKTHROUGH/STRONG；跨域迁移必须写 source/target domain、 可迁移机制与 mismatch risk。A+B 没有机制 delta/判别预测 = critical fail，不准送 idea-critique。 - 防伪多样:
candidate_dedup.py(接_shared/semantic_sim)两两算相似,批内 mean+1σ 自动标"疑似换皮变体对" → 合并或重发散,别拿同一 idea 的变体凑数。 - 撞车前置自查 + 产 findings:
idea_selfcheck.py --domain-map <literature-search 的 --json-out>对每个候选用semantic_sim找最像的前作 + facet 槽位 → 产light.findings.v1(撞车/伪多样/覆盖,warn)→ 交总控run_checkpoint --stage 3聚合。 - 立项卡完整性门:每条候选填立项卡(
templates/idea_card.md)→card_gate.py校验 必填非空 + 非敷衍占位 + 最近邻≥3 带检索留痕 + 新颖性归三档(残卡/敷衍 exit 1 拦下,交 idea-critique 前过); ★Round 2 R1 加可证伪 warn:「最小验证实验」「失效条件」缺可测量阈值/量化失效条件 → 警示(借 K-Dense testability + Galaxy-Dawn falsification,只 warn 不阻断,真判归 idea-critique)。 - 分层排序:
rank_ideas.py分 moonshot/solid/safe 三道各自排序再 round-robin(突破口不被性价比压杀);swiss_rank.py瑞士轮 ELO 两两配对(压过自报绝对分,Si 实测自评一致性仅 ~53%)。
What ships with it
18 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.
- .gitignore 108 B
- examples/candidates.example.json 1.5 KB
- examples/idea_candidates.example.md 8.3 KB
- idea-resource-map.md 13 KB
- references.md 19 KB
- scripts/candidate_dedup.py 8.3 KB runs code
- scripts/card_gate.py 20 KB runs code
- scripts/gap_evidence_gate.py 22 KB runs code
- scripts/idea_genealogy.py 21 KB runs code
- scripts/idea_selfcheck.py 21 KB runs code
- scripts/innovation_engine.py 19 KB runs code
- scripts/provocation_gen.py 11 KB runs code
- scripts/rank_ideas.py 10 KB runs code
- scripts/swiss_rank.py 8.5 KB runs code
- templates/idea_card.md 4.3 KB
- templates/idea-gap-evidence.example.json 620 B
- templates/idea-genealogy.example.json 2.2 KB
- templates/innovation-engine.example.json 2.1 KB
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 · 328 lines · 401 tokens per session scan A 981b89aee8e9
light-idea-generation is a skill published in the GitHub repository Light0305/Light-skills (617 stars, last pushed 2mo ago), licensed MIT. It adds 401 tokens to every session and 8,988 once invoked, about $0.0020 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.
anti-defensive-writing-en
Stops defensive writing across the entire paper lifecycle — writing, revising, cutting, and organizing experiments. Treats the paper as a press conference, not a project summary, lab log, or self-audit: identify the single most publishable strength of the work and build the most favorable, complete, and persuasive…
anti-defensive-writing
A Chinese-language writing guide for presenting a research paper around its strongest supported contribution. It treats the paper as a focused academic presentation rather than a project diary or complete lab record.
paper-writing
Research paper writing assistant that enforces Arpit Gupta's editorial principles, voice profile, and writing workflow. MANDATORY TRIGGERS: Use this skill whenever the user mentions writing a paper, drafting a section, revising a section, editing a paper, reviewing a draft, rewriting an introduction, writing an…
research-writing
A collection of 30 prompt templates for writing and reviewing scientific papers. It covers tasks such as translating, editing, summarizing research, writing sections, creating figure captions, and preparing reviewer replies.
academic-paper-writing-skill
Evidence-first academic research workflow for topic ideation, scholarly search, paper reading, literature and systematic reviews, study and experiment design, statistics and data analysis, scientific figures, manuscript drafting and polishing, citation checks, peer review, rebuttals, submission packages, theses, and…