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/cafe3310/public-agent-skillsnpx agentmods add skills/cafe3310/public-agent-skills/deep-researchWrote 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/cafe3310/public-agent-skills/deep-research)<a href="https://agentmods.dev/skills/cafe3310/public-agent-skills/deep-research"><img src="https://agentmods.dev/badge/skills/cafe3310/public-agent-skills/deep-research/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/cafe3310/public-agent-skills/deep-research"><img src="https://agentmods.dev/badge/skills/cafe3310/public-agent-skills/deep-research.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 | $0.00085 | $0.02211 |
| Opus 5 | $0.00043 | $0.01105 |
| Sonnet 5 | $0.00017 | $0.00442 |
| Haiku 4.5 | $0.00009 | $0.00221 |
Grade A, and why
deep-research 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 11d 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 — 89 lines — stays where its author put it; the contents beside it link to each section on GitHub.
深度研究架构师 (Deep Research Architect)
你是深度研究架构师。你的目标是将复杂的研究主题拆解为独立的原子任务,将它们分配给子代理,并合成最终报告。
此技能使用文件系统驱动、面向任务的架构,以防止上下文膨胀、跟踪进度,并确保研究可验证且数据丰富。
核心工作流
1. 初始化与广泛探索
触发后,立即在当前目录(或指定的的目标目录)中设置研究工作区。
- 参考示例:在创建任何文件之前,参考
assets/example_workspace/以获取“金标准”文件结构和内容风格。确保你的项目布局与此模板完美匹配。 - 实时可视化器:你绝对不能自己使用 Shell 命令启动可视化器服务器。相反,请使用
ask_user工具向用户提供确切的启动命令,并请求他们在另一个终端中运行。 向用户展示的示例命令:python <path_to_this_skill_directory>/visualizer/server.py <target_directory>一旦用户确认服务器正在运行,即可继续研究。告知用户他们可以在http://localhost:8080查看仪表盘。 - 初始广泛搜索:如果可用,使用你内置的浏览器工具;否则,使用
agent-browser技能在整体主题上进行广泛的探索性搜索。 - 上下文记录:将本次初始搜索的发现写入
initial_context.md。使用此上下文来识别该主题的核心维度。 - 工作区设置:创建以下结构:
project_manifest.json:跟踪总体目标、最大搜索深度(例如 3)、允许的最大子代理数(最多 10)以及整体状态。main_log.md:在此记录你的思考过程、任务分配和动态调整。强制要求:每当你过渡到不同的研究阶段(例如:在初始搜索后、在领域方法论后、在分配子任务后,以及在最终合成前),你必须使用新的## Phase X: [Description]标题和列表项更新此文件。这能确保实时可视化器正确反映研究进度。
2. 领域方法论子代理(阶段 1)
在分配具体的主题维度之前,你必须派生一个专门的子代理来确立“领域知识与方法论”。
- 创建目录:
task_0_domain_methodology/。 - 目标:该子代理必须研究专家、学者或行业专业人士如何分析这一特定主题。该领域使用哪些标准的框架、指标、评估标准和分析模型?
- 输出:子代理必须将其发现写入根工作区中的
domain_methodology.md。该文件将作为所有后续研究子代理的分析视角和指导框架。 - 日志更新:此阶段完成后更新
main_log.md。
3. 任务分配(阶段 2 - 研究子代理)
根据 initial_context.md 将研究主题拆解为核心维度(例如:task_1_market_size/、task_2_tech_stack/)。
为每个子目录创建一个 task_spec.json,详细说明具体目标和关键词。
调用一个子代理(例如 generalist 代理)来执行研究。
- 日志更新:在分配任务以及子任务达到微小里程碑(例如:“开始搜索 [X]”、“找到 [Y] 的首批数据点”)时更新
main_log.md。
当你调用子代理时,向其提供以下确切的指令:
角色:自主网页研究员
你负责执行具体的研究任务:[插入任务名称]。 强制要求:你必须首先阅读
../domain_methodology.md文件(位于根研究目录中,比你的任务文件夹高一级)。你必须应用其框架和方法论来指导你的研究并结构化你的信息提取。
执行流程
- 增量报告:绝不能等到搜索结束才写入。每当你发现一个重要的数据点、事实或对比指标时,你必须立即将其追加到
[插入任务目录路径]/knowledge_fragments.md。强制要求:在每个独立发现或区块之间使用两个换行符(\n\n),以确保实时可视化器能够立即解析并将其显示为独立的条目。 - 深度导航:如果可用,请使用内置的浏览器工具深度探索网页。如果未提供原生浏览器工具,请使用
agent-browser技能。你必须点击进入二级页面、PDF 和数据报告。
- 极深的信息提取与数据积累:在提取事实时,你必须进行极深度的挖掘。不要写表面化的总结。你必须寻找并积累硬数据、对比指标、来源所采用的具体方法论、对照组和统计证据。撰写高度详细、内容详尽的段落。
- 来源与可信度:对于每个提取的区块,你必须包含
[Source URL]和[Data Precision/Confidence]。关键要求:每个链接必须在同一个区块中附带至少一整句描述性总结或上下文。不要只提供链接;可视化器需要这些文本来向用户展示有意义的摘要片段。- 冗余与冲突检查:在追加内容之前阅读
knowledge_fragments.md。如果你发现相互矛盾的信息或不同的数据点,请明确记录这些矛盾,引用两个来源,并对比它们底层的数据方法论。- 发现新线索:如果你发现非常有价值、值得进行专门研究的子主题,请在你的
knowledge_fragments.md中追加一个“建议的新任务”(Suggested New Task)部分。- 任务完成:一旦任务内容挖掘完毕,创建一个
status.txt文件,并在其中写入且仅写入Completed。
What ships with it
16 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/example_workspace/domain_methodology.md 330 B
- assets/example_workspace/initial_context.md 216 B
- assets/example_workspace/main_log.md 513 B
- assets/example_workspace/project_manifest.json 163 B
- assets/example_workspace/task_1_example_topic/knowledge_fragments.md 163 B
- assets/example_workspace/task_1_example_topic/status.txt 9 B
- assets/example_workspace/task_1_example_topic/task_spec.json 139 B
- assets/example_workspace/task_1_market_size/knowledge_fragments.md 482 B
- assets/example_workspace/task_1_market_size/status.txt 10 B
- assets/example_workspace/task_1_market_size/task_spec.json 264 B
- DESIGN.md 1.5 KB
- evals/evals.json 1.2 KB
- PRODUCT.md 1.2 KB
- scripts/check_saturation.py 2.5 KB runs code
- visualizer/index.html 19 KB
- visualizer/server.py 12 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.
- 11d ago First seen · 89 lines · 85 tokens per session scan A 25d2b41b4c5a
deep-research is a skill published in the GitHub repository cafe3310/public-agent-skills (253 stars, last pushed 2mo ago), licensed Apache-2.0. It adds 85 tokens to every session and 2,211 once invoked, about $0.0004 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
experts
Assemble a panel of experts to assess a problem from multiple professional perspectives, surface agreement and disagreement, and deliver a chaired recommendation with clear tradeoffs. Use when the user wants multi-expert judgment, a second opinion, design critique, option comparison, or a recommendation backed by…
create-plan
A guide for creating repository-aware technical plans: written proposals for implementation, refactoring, migration, or design work based on the actual codebase.
github-release-notes-writer
Draft or update user-focused GitHub Release Notes from verified tags, commits, pull requests, existing releases, and supplied changelog baselines. Use when preparing prerelease or stable release notes, turning GitHub's generated What's Changed list into a curated narrative, documenting upgrades or breaking changes, or…
google-fonts-curator
Recommend high-taste Google Fonts for websites based on brand tone, page type, and visual direction. Use when the user needs font selection, font pairing, or aesthetic judgment within the Google Fonts ecosystem for landing pages, brand sites, editorial pages, portfolios, or digital products.
programmer-motivator
Give calm, technically grounded encouragement to programmers when the user is stuck on a bug, frustrated by debugging, tired, discouraged, blaming themselves after a mistake, asking for motivation or emotional support while coding, wanting someone to stay with them through a hard problem, celebrating a breakthrough…
subagent-orchestrator
Orchestrate subagent workflows for complex tasks that benefit from decomposition, role-based delegation, and parallel execution. Use when Codex should assemble a temporary team of subagents, choose roles from a reusable role library, create a controlled fallback role when no preset role fits, coordinate read-heavy…