Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add zj-unicom-ai/UniEmployee --skill industry-deep-researchgit clone --depth 1 https://github.com/zj-unicom-ai/UniEmployeeWrote 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/zj-unicom-ai/uniemployee/industry-deep-research)<a href="https://agentmods.dev/skills/zj-unicom-ai/uniemployee/industry-deep-research"><img src="https://agentmods.dev/badge/skills/zj-unicom-ai/uniemployee/industry-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/zj-unicom-ai/uniemployee/industry-deep-research"><img src="https://agentmods.dev/badge/skills/zj-unicom-ai/uniemployee/industry-deep-research.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00041 | $0.02059 |
| Opus 5.5 | $0.00016 | $0.00824 |
| Sonnet 5.5 | $0.00008 | $0.00412 |
| Haiku 4.5 | $0.00004 | $0.00206 |
Grade A, and why
industry-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 yesterday.
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 — 123 lines — stays where its author put it; the contents beside it link to each section on GitHub.
行业深度研究
接到行业研究任务时,先按本规程拆解问题,再检索和综合证据。研究目标是解释行业如何运作、正在发生什么变化、证据支持哪些判断,以及哪些问题仍不确定。
适用范围
- 行业全景、市场规模与增长驱动
- 产业链结构、关键环节、成本与价值分布
- 竞争格局、代表企业和商业模式
- 技术路线、产业化进展和采用障碍
- 政策法规、标准、贸易与供应链环境
- 中长期趋势、情景推演、机会和风险
简报、单条消息核实和单个竞品对比优先使用对应的专门技能;需要时可在深度研究中引用这些技能的分析方法。
研究步骤
1. 明确研究口径
记录或说明:
- 研究对象和行业边界
- 地理范围
- 时间范围和基准日期
- 用户希望支持的判断或决策
- 采用的假设、暂不覆盖的内容
用户没有指定时,围绕用户提到的行业开展研究;小察工作台首批持续观察主题为具身智能、AI 前沿观察和政策分析,也覆盖智能硬件及其上下游和关联行业。
2. 建立研究问题
根据任务选择相关问题,不必机械覆盖全部维度:
- 需求来自哪些客户、场景和预算?需求是否在增长?
- 市场规模数字采用什么定义、地区、年份和计量单位?
- 价值链有哪些环节,利润、议价权、进入壁垒和依赖关系如何分布?
- 主要参与者分别处于什么位置,竞争优势由什么形成?
- 哪些技术路线在成熟、验证或规模化阶段?成本和采用障碍是什么?
- 哪些政策、标准和供应链变化会影响行业,何时生效?
- 未来 1–3 年哪些变量最可能改变行业判断?如何观察?
3. 检索与记录证据
- 使用具体关键词,围绕不同研究问题分别检索。必要时按企业公告、行业数据、政策、技术和应用场景分组。
- 优先使用政府部门、监管机构、行业组织、公司公告/财报、标准文件和原始研究。媒体报道与二手研究用于补充和发现线索。
- 关键结论尽量由两个相互独立的来源支持。需要多轮采集时委派
intel-scouter,并要求其返回链接、发布日和原文要点。 - 每条证据记录来源名称、链接、发布日期、数据所描述的时期及其口径。
- 页面无法访问、搜索结果不含原文或来源之间互相矛盾时,明确记录限制,不推测缺失内容。
4. 核验与分析
- 将来源披露、第三方估算、分析判断和情景假设分开。
- 核对数字的地域、统计范围、计量单位、价格口径、样本和时间窗口。
- 比较冲突信息时,先检查定义和方法差异,再判断是否可以并列;无法对齐时不强行求平均。
- 解释证据之间的联系:需求如何传导到产业链,技术或政策如何改变成本、竞争位置和商业化节奏。
- 不把相关性写成因果,不从个别企业案例推断全行业普遍规律。
- 研究动态信息时写明检索截止日;较早的资料仍可用于结构性判断,但要标明资料时间。
5. 形成判断
明确回答用户的问题,并列出:
- 证据充分的核心发现
- 对目标企业、产品或决策的可能含义
- 关键不确定性和证据缺口
- 可继续跟踪的指标、事件或验证条件
若提供情景推演,写明驱动因素、假设和触发信号,不将情景当成预测事实。
默认交付结构
按任务裁剪,不为满足模板加入空泛章节:
- 结论摘要:直接回答研究问题,列出最重要的发现。
- 研究口径:对象、范围、时间、方法和资料截止日。
- 行业结构:需求、产业链、竞争、技术和政策中与问题相关的内容。
- 关键判断:说明证据如何支持判断,并标记事实、估算或推断。
- 机会、风险与不确定性:说明影响路径和适用条件。
- 观察清单:建议跟踪的指标、来源和触发信号。
- 来源清单:标题、发布机构、发布日期和可访问链接。
质量检查
输出前检查:
- 是否覆盖了用户真正要判断的问题?
- 核心数字是否带口径、时间、单位和来源?
- 来源是否可访问,发布日期是否准确?
- 事实、估算、分析和建议是否清楚区分?
- 是否指出了来源冲突、资料缺口和结论限制?
- 建议是否能由前面的证据推出?
写入小察研究工作台
当研究结果适合成为某个主题的持续观察时,调用 save_market_observation 保存一条记录:
topic_id只能使用embodied-intelligence、ai-frontier、policy-analysis。- 每条资讯必须且只能选一个所属细分领域,写入对应的
category_id;不要多选,也不要只凭标题猜分类。分类按主要研究信号确定:- 具身智能:
home-service-robots(家庭服务机器人)、industrial-robots(工业机器人)、commercial-service-robots(商业服务机器人)、robot-platforms(机器人本体与核心部件)、embodied-models(具身模型与训练)、robot-commercialization(量产与商业化)。 - AI 前沿观察:
foundation-models(基础模型)、multimodal-ai(多模态 AI)、ai-agents(智能体与工具链)、compute-infrastructure(算力与基础设施)、ai-applications(产品与行业应用)、ai-ecosystem(企业与开源生态)。 - 政策分析:
national-strategy(国家战略与规划)、industry-support(产业扶持政策)、regulation-compliance(监管与合规)、standards-certification(标准与认证)、local-pilots(地方试点与项目)、policy-impact(政策落地与产业影响)。 如果无法从来源确认主要归属,继续核对;仍无法判断时不要入库,并说明缺少的证据。
- 具身智能:
- 标题简明描述信号;摘要说明发生了什么及其关注价值;正文保留口径、证据、事实与判断、不确定性和后续观察项。
sources_json使用 JSON 列表,逐项提供标题、URL、发布日期和发布机构。不要把搜索摘要当作已核实的原文证据。- 简单问答或一次性咨询不必保存。只有可复用、可追踪的行业信号才写入工作台。
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.
- yesterday First seen · 123 lines · 41 tokens per session scan A f6d56f5791e2
industry-deep-research is a skill published in the GitHub repository zj-unicom-ai/UniEmployee (615 stars, last pushed yesterday), licensed MIT. It adds 41 tokens to every session and 2,059 once invoked, about $0.0002 per session on Opus 5.5. 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 2026-10-27.
Other skills, from other repositories
action-contract-execution-feedback-loop
A specification method for making a business action into a tool an AI agent can call safely. It defines the action’s inputs, outputs, permissions, preconditions, expected effects, error handling, and audit information.
fact-reason-action-business-loop
A method for connecting observed facts, business rules, and permitted actions into one traceable loop. It records what happened, explains why a conclusion follows, defines allowed actions, and captures the results.
five-ring-ontology-engineering-lifecycle
A five-stage method for planning an enterprise ontology, a structured model of the concepts and relationships used across a business, from source material through ongoing use.
ontology-ai-application-pattern-selection
A method for choosing how an AI system should handle a business problem after deciding that an ontology—a structured map of concepts and relationships—is suitable. It compares patterns such as workflow automation, decision support, knowledge answering, collaboration, and continuous planning.
ontology-ai-scenario-fit-and-spike
A method for deciding whether an ontology is suitable for an enterprise AI scenario and testing that choice with a small end-to-end sample. An ontology is a structured model of concepts, rules, and relationships.
ontology-constraint-and-knowledge-injection
A method for deciding how an ontology’s knowledge should reach an AI model: directly in its instructions, through RAG, or through fine-tuning. RAG retrieves relevant information at answer time; fine-tuning changes the model using training examples.