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 agentmods add commands/hayes-zhang/deep-research/deepgit clone --depth 1 https://github.com/Hayes-Zhang/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/commands/hayes-zhang/deep-research/deep)<a href="https://agentmods.dev/commands/hayes-zhang/deep-research/deep"><img src="https://agentmods.dev/badge/commands/hayes-zhang/deep-research/deep.svg" alt="Measured on agentmods" 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 | $0.00027 | $0.06210 |
| Opus 5 | $0.00014 | $0.03105 |
| Sonnet 5 | $0.00005 | $0.01242 |
| Haiku 4.5 | $0.00003 | $0.00621 |
Grade A, and why
deep scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 493 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/deep — 多视角深度研究
⚠️ 执行模式:Agent Team(强制) 本命令必须使用 Agent Team 模式执行,禁止退化为 Subagent 模式。
- ✅ 必须调用 TeamCreate 创建团队
- ✅ 必须使用 Task 的 team_name + name 参数生成 teammate
- ✅ teammate 之间必须通过 SendMessage 协作
- ❌ 禁止使用纯 Subagent(即不带 team_name 的 Task 调用)
- ❌ 禁止跳过 TeamCreate 直接启动 researcher
你将对以下主题进行 7 个维度的深度协作研究:
研究主题:$ARGUMENTS
阶段 1:理解需求
首先,分析用户提供的研究主题 $ARGUMENTS。
- 如果主题足够明确,直接进入阶段 2
- 如果主题模糊或过于宽泛,使用 AskUserQuestion 工具追问:
- 你具体想了解什么方面?
- 你的研究目的是什么?(产品设计 / 技术选型 / 竞品分析 / 其他)
- 有没有特别关注的角度?
将主题分解为适合 7 个视角分别研究的子问题。
阶段 2:创建团队 & 分配任务
2.1 创建研究团队
使用 TeamCreate 创建一个研究团队:
TeamCreate:
team_name: "deep-research-{timestamp}"
description: "多视角深度研究:[研究主题摘要]"
其中 {timestamp} 使用当前时间戳(如 20260206-143000),确保团队名称唯一。
验证:TeamCreate 调用成功后,确认团队名称已创建,再继续下一步。如果 TeamCreate 失败,停止执行并报告错误。
2.2 创建 8 个任务(7 研究任务 + 1 评审任务)
使用 TaskCreate 为 7 个研究视角和 1 个 critic 评审各创建一个任务:
| # | 任务名称 | 描述 |
|---|---|---|
| 1 | 产品设计与PM视角研究 | 从产品设计与PM角度研究 [主题] |
| 2 | 交互设计视角研究 | 从交互设计角度研究 [主题] |
| 3 | 技术方案视角研究 | 从技术方案角度研究 [主题] |
| 4 | 行业最佳实践视角研究 | 从行业最佳实践角度研究 [主题] |
| 5 | 学术研究视角研究 | 从学术研究角度研究 [主题] |
| 6 | 历史科技产品视角研究 | 从历史科技产品角度研究 [主题] |
| 7 | 社交媒体观察视角研究 | 从社交媒体观察角度研究 [主题] |
| 8 | 对抗性同行评审 | 在 7 份 perspective 报告完成后,审查并产出 Issue Report |
2.3 并行启动 8 个 teammate(7 researcher + 1 critic)
重要:必须在一条消息中并行发起所有 8 个 Task 调用(7 个 researcher + 1 个 critic),不要串行启动。critic 启动后会进入 idle 状态等待你在阶段 3.4 发起 review 请求——这样做能节省一次单独的 TaskCreate。
每个 Task 调用必须包含以下参数(缺一不可):
subagent_type: 插件 agent 名称(如"deep-research:researcher-product")team_name: 与 2.1 创建的团队名称完全一致name: teammate 名称(如"product")prompt: 研究任务描述(见下方模板)model:sonnetmode:"bypassPermissions"(加速执行,减少权限弹窗)description: 简短描述
⚠️ 禁止事项:
- 不得省略
team_name参数(否则会退化为 Subagent,失去团队协作能力) - 不得省略
name参数(否则 teammate 无法被识别和通信) - 不得串行启动(8 个 Task 调用必须在一条消息中并行发起)
critic 的 prompt 模板(与 7 个 researcher 的 prompt 不同——critic 不做研究,启动后等待 review 请求):
Task tool:
subagent_type: "deep-research:critic-reviewer"
team_name: "deep-research-{timestamp}"
name: "critic"
mode: "bypassPermissions"
prompt: |
## 你的任务
**团队名称**:deep-research-{timestamp}
**你的角色**:critic-reviewer(对抗性同行评审)
**用户原始研究主题**:[用户的研究主题]
**用户问题的语言**:[zh 或 en,用于决定 Issue Report 的输出语言]
你是质量门,不做原始研究。等 7 个 researcher 完成报告后,team lead 会通过 SendMessage 发起 review 请求,附上 7 份 perspective 报告的路径。
**现在的工作**:
1. 用 TaskGet 读取你的任务详情
2. **解析 lead 的名字(强制)**:读取 ~/.claude/teams/{团队名称}/config.json,记住 lead 的 name 字段,后续 SendMessage 的 recipient 必须用它
3. 用 TaskUpdate 将任务标记为 in_progress(让 lead 知道你在线)
4. 进入 idle 等待状态,**不要主动开始任何 WebSearch / WebFetch**
5. 收到 lead 的 review 请求后,按你的 system prompt 中的 Review Checklist 工作,产出 Issue Report
6. 把 Issue Report 通过 SendMessage 发给 lead
7. **立即** TaskUpdate 标记为 completed(不要等 shutdown_request)—— 否则 lead 会以为你卡死
**预计被触发的时间**:7 个 researcher 完成后约 10-15 分钟。
description: "Critic Review(质量门)"
model: sonnet
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.
- 5d ago First seen · 493 lines · 27 tokens per session scan A eea3e51dc015
deep is a command published in the GitHub repository Hayes-Zhang/deep-research (4 stars, last pushed 3mo ago), licensed MIT. It adds 27 tokens to every session and 6,210 once invoked, about $0.0001 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 2026-08-31.
Other commands, from other repositories
notebook
Local NotebookLM over a FOLDER of documents using Antigravity (agy). Sweeps each document (PDF with text, scanned PDF, image, docx) into an objective-driven Markdown summary, then builds a relevance INDEX and a cited master summary. Incremental cache (re-runs only re-summarize changed docs / changed objective) and…
deep-research
Deep, multi-source, fact-checked web research with agy — reach for it when a decision or design depends on getting it right and a single-shot answer is not enough (architecture / tool / vendor choices, thorough landscape scans, anything you will act on). Builds an evidence matrix + a plan you approve, then agy browses…
report
Generate a publication-grade HTML report from a markdown source using the TRAID Design System catalog (5 templates). Agy matches content to template, you confirm, agy produces final branded HTML.
notebook-query
Query the notebook knowledge base (SQLite) built by /agy:notebook — precise, grounded, cited. Ask in natural language ("sum the amounts by category", "which docs mention 'Acme Corp'", "build a project timeline") or pass raw SQL. Read-only. Use this when you need exact aggregates/lookups across a document corpus…
graph
Build a knowledge GRAPH of a folder (code + docs) with Graphify — tree-sitter ASTs + NetworkX + Leiden communities + an interactive graph.html. The code graph is built LOCALLY and costs zero tokens on any assistant; Gemini via agy only names the communities. Then Claude reads graph.json / GRAPHREPORT.md to reason…
model
Show or set the agy (Antigravity / Gemini) model by writing /.gemini/antigravity-cli/settings.json — the reliable way to switch models (the unreliable --model flag silently falls back to the default). No agy call; takes effect immediately, no plugin reinstall.