Docs & Knowledge · daymade/claude-code-skills
deep-research
Generate format-controlled research reports with evidence tracking, citations, source governance, and multi-pass synthesis. This skill should be used when users request a research report, literature review, market or industry analysis, competitive landscape, policy or technical brief. Triggers: "帮我调研一下", "深度研究", "综述报告", "深入分析", "research this topic", "write a report on", "survey the literature on", "competitive analysis of", "技术选型分析", "竞品研究", "政策分析", "行业报告". V6 adds: source-type governance, AS_OF freshness checks, mandatory counter-review, and citation registry. V6.1 adds: source accessibility (circular verification forbidden, exclusive advantage encouraged).
L3 Networked · Declare domain & key scopeV1 Commit-PinnedNetwork AccessWrite FileDual-pass Audit
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.
1How It Works & Mechanism
This is a pure prompt-orchestrated skill: containing no scripts or executable code, it drives the host LLM to generate research reports via a "Lead Agent + Parallel Subagent" workflow alongside 14 markdown knowledge files.
deep-research/ 目录结构（ls）+ references/ 14 个 .md
SKILL.md 23.2KB + references/（completeness_review_checklist.md, counter_review_team_guide.md, enterprise_analysis_frameworks.md, enterprise_quality_checklist.md, enterprise_research_methodology.md, formatting_rules.md, quality_gates.md, report_template_v6.md, research_notes_format.md, research_plan_checklist.md, research_report_template.md, source_accessibility_policy.md, source_quality_rubric.md, subagent_prompt.md）——无 scripts/ 目录、无任何脚本文件
All capabilities stem from instruction design, and execution relies entirely on the host environment providing web_search/web_fetch/subagent/file writing.
Context isolation is the core mechanism: subagents search and crawl independently, writing only distilled notes into research-notes/task-{id}.md, while the lead agent reads only the notes rather than raw search results.
SKILL.md:16-39
Dispatch ──→ Subagent A ──→ writes task-a.md ──┐ ... research-notes/ <──┘ | Context efficiency: Subagents' raw search results stay in their context and are discarded. Lead agent sees only distilled notes (~60-70% context reduction).
The file system simultaneously serves as a communication bus among subagents and between subagents and the main agent; the note format acts as the interface protocol.
Note files serve as the sole traceability channel where every fact must trace back to a specific note line, establishing the first lock against hallucinations.
references/research_notes_format.md:3-5
The research notes are the ONLY communication channel between subagents and the lead agent. Every fact in the final report must be traceable to a line in these notes. No exceptions.
Combined with P7 Verify's [n] registry verification and 5+ spot checks, "traceability" is turned into operational procedure rather than just a slogan.
P3 constructs a global citation registry and applies quantitative quality gates; sub-standard sources are explicitly dropped and never reused.
SKILL.md:283-305
Standard: ≥12 approved sources, ≥5 unique domains, ≥30% official | Lightweight: ≥6 approved sources, ≥3 unique domains, ≥20% official | Max single-source share: ≤25% (Standard), ≤30% (Lightweight) ... **Critical rule:** These [n] are FINAL. P5 may only cite from Approved list. Dropped sources never reappear.
Source type (official/academic/secondary-industry/journalism/community/other), as_of, and authority 1-10 are tagged individually, allowing "source quality" to be quantitatively inspected.
V6.1's "source accessibility" governance distinguishes between public / semi-public / exclusive-user-provided / private-user-owned, prohibiting the use of user private data for self-verification (circular verification) while encouraging user-paid sources for third-party research.
SKILL.md:307-317
When researching the user's own company/assets, if you discover data in user's private accounts (e.g., user's domain registrar showing they own domains), you MUST: 1. Reject it from the registry... 4. Report from external investigator perspective only | When user EXPLICITLY PROVIDES their paid subscriptions or private APIs for third-party research (e.g., "Use my Crunchbase Pro to research competitors"), you SHOULD: 1. Accept it as "exclusive-user-provided" accessibility
A rule derived from a real case involving a ByteDance subsidiary: inspecting the user's own company must rely solely on public sources, whereas inspecting competitors can leverage user-provided exclusive sources such as Crunchbase Pro or PitchBook.
P6 enforces mandatory counter-review: every primary conclusion must undergo counter-interpretation checks, and the search is restarted if fewer than 3 issues are identified.
SKILL.md:404-412
1. **Could the conclusion be wrong?** 2. **Which high-impact claims depend on a single source?** 3. **Which claims lack official/academic support?** 4. **Are stale sources used for time-sensitive claims?** 5. **Find ≥3 issues** (re-examine if 0 found)
The ideal workflow dispatches 4 named agents (claim-validator/source-diversity-checker/recency-validator/contradiction-finder) + a coordinator for parallel review; it falls back to a manual checklist if unavailable.
Enterprise mode provides a comprehensive company research pipeline: 6-dimensional data collection → SWOT/weighted competitive barrier scoring/risk matrix/composite scorecard → L1/L2/L3 three-tier quality control.
SKILL.md:130-141 + references/enterprise_research_methodology.md
E2: Six-dimension data collection... D1: Company fundamentals ... D6: Internal/proprietary sources ... E3: Structured analysis frameworks: SWOT analysis (evidence-backed, 4 quadrants × 3-5 entries); Competitive barrier quantification (7 dimensions, weighted score); Risk matrix (8 categories, probability × impact); Comprehensive scorecard (6 dimensions, weighted total)
Barrier scoring spans 7 dimensions each with weights and scoring anchors (e.g., Network Effects 20%), 8 risk categories must be evaluated, and every cell requires evidence and sources.
The wrap-up stage routes outputs to sibling skills within the same marketplace: fact-checker for review, ppt-creator for slides, and pdf-creator for export.
SKILL.md:539-543
A) Verify facts — run /fact-checker on the report (Recommended) B) Create slides — run /daymade-docs:ppt-creator from the findings C) Export as PDF — run /daymade-docs:pdf-creator for formal delivery
Within the same monorepo, fact-checker/ as well as daymade-docs/ppt-creator and daymade-docs/pdf-creator indeed exist (directories verified), representing an in-ecosystem referral rather than an external service.
2Core Capabilities
01Mode-adaptive research planning: Topic (Enterprise/General) × Depth (Standard 3000-8000 words across 5-6 tasks / Lightweight 2000-4000 words across 3-4 tasks)
02Parallel task board: each task includes expert persona/objectives/preset queries/depth (DEEP crawls 2-3 full texts or SCAN), Group A/B dependency dispatch, maximum 3 concurrency
03Subagent research notes: strictly formatted with Sources (real URL + Source-Type + As Of + Authority), ≤10 Findings, Deep Read Notes, and Gaps
04Citation registry and quality gates: approved/dropped partitioning, domain count and official source ratio thresholds, single-source cap, deduplication, and authority score re-evaluation
05AS_OF freshness discipline: P0 establishes the research baseline date, time-sensitive statements require source publication dates, and outdated sources have confidence downgraded and are flagged in the registry
06Multi-pass drafting + mandatory counter-review + final verification: P5 section-by-section drafting (confidence tag + counter-explanation per section) → P6 identifying ≥3 issues → P7 full [n] verification and 5+ spot checks
07Enterprise-grade company research (6-dimensional collection / framework quantification / L1-L3 QA / 7-chapter template + appendix), tailored for due diligence and competitive analysis
08Information black-box protocol: subjects with zero public footprint are honestly reported as NO PUBLIC INFORMATION AVAILABLE, and using user private credentials to fill gaps is strictly prohibited
3External Dependencies
| Kind | Dependency / Endpoint |
|---|---|
| cli | claude（Claude Code CLI，headless -p 派子代理） |
| cli | 宿主 agent 内建工具 web_search / web_fetch（技能本身不实现） |
| package | Cowork / DeerFlow / OpenClaw（可选子代理调度宿主，纯说明性，非依赖） |
| api | 无硬编码 API；流程允许用户在会话中提供 Crunchbase Pro / PitchBook / Wind 等付费数据源给模型用于第三方研究 |
| network | 无固定网络端点；全部 URL 仅出现在示例格式块（nationalgeographic.com / cam.ac.uk / wikipedia.org / quora.com / study.com / blogs.microsoft.com / effectivealtruism.org），非代码调用点 |
| package | 同仓库兄弟 skill 转介：fact-checker、daymade-docs:ppt-creator、daymade-docs:pdf-creator（已核实目录存在） |
4Recommended Governance & Evidence
Suggested Controls for Safe Operation
For Agent Runtime
Declare allowed domains and key scope
For Administrators
Least privilege keys, retain request logs
Evidence & Triggers (File Paths & Excerpts)
- 网络端点: Declared external network access: host web_search/web_fetch: accesses arbitrary public websites based on user research topics (targets cannot be pre-enumerated, an essential behavior of web research; the skill itself has no curl/fetch/
- 本地工作目录: File read/write and artifact output: workspace/research-notes/task-{id}.md, registry.md, counter-review-inputs/ (dr
Audit Notes: Contains no scripts, no credential reads, and no hardcoded endpoints (characteristics of Green/Blue tiers); however, web search + full-text crawling are its core capabilities. Outbound targets are determined by the user's research topic and cannot be pre-enumerated, and crawling content back into the model context presents a prompt injection surface—classifying it as a 'routine web research invocation, use with awareness' tier. The workflow allows the model to use user-provided paid data sources (e.g., Crunchbase Pro) in the session for third-party research, which constitutes voluntary user authorization rather than credential theft.
5Second-Pass Verification
- [ok] 外部 CLI 依赖 claude -p（--allowedTools web_search,web_fetch,write） — references/subagent_prompt.md:91-99 真实存在，含单任务与并行 for 循环写法
- [ok] web_search/web_fetch 为功能前提 — SKILL.md:110-111 P0 能力表，缺失即 Stop / SCAN-only；另有 references/subagent_prompt.md:23-24 指令
- [ok] Cowork / DeerFlow / OpenClaw 派发途径 — references/subagent_prompt.md:104-115 存在说明与 task() 示例；但仅为宿主选项说明，无实现代码
- [ok] 跨 skill 转介 /fact-checker、/daymade-docs:ppt-creator、/daymade-docs:pdf-creator — SKILL.md:540-542 存在；fact-checker/、daymade-docs/ppt-creator/、daymade-docs/pdf-creator/ 目录在同仓库均已核实存在
- [ok] 无脚本/无可执行代码 — deep-research/ 仅 SKILL.md + .security-scan-passed + references/ 14 个 .md，无 scripts/、无 .py/.sh/.mjs
- [ok] 无凭证读取/无密钥硬编码 — 全目录正则扫 api[_-]?key|Bearer|sk-|password|Authorization|secret 无真实命中（仅 task-a.md 中 'sk-a' 子串误报）；.security-scan-passed 亦由 gitleaks 扫描背书
- [ok] 无硬编码网络端点/无 HTTP 调用代码 — 全部 URL 位于示例数据块（research_notes_format.md:32-35、122-131 等），无 curl/wget/requests/axios/fetch
- [ok] SKILL P3 质量门数值与 quality_gates.md Gate2 一致 — 两处均为 ≥12/≥6 源、≥5/≥3 域名、≥30%/≥20% official、单源 ≤25%/≤30%，无矛盾
6Verdict & Scenarios
Best for: Users who need traceable, multi-source cross-validated, structured research reports with citations and confidence tags: market/industry/policy/tech stack analysis, competitive and due diligence company research, literature reviews; especially suitable for Claude Code/hosts supporting subagent dispatch and web tools, as well as research users who prioritize hallucination prevention and citation standards.
Not for: Scenarios requiring script-level automation, direct API batch scraping, or deterministic search engines (this skill is zero-code and relies entirely on host web tools + LLM execution); one-line quick Q&A queries (the workflow is too heavy, typically involving 3-8 tasks and multi-pass drafting); environments without web_search/web_fetch or with network access disabled (triggers Stop at P0).
Install Agent-Ready
① Verified Mirror 2026-09-06
Option A · Direct DownloadDownload deep-research.tar.gz
sha256:
Option B · Copy Guide for Agentcddc8d3acbe77673…Agents automatically download and verify mirrors from this JSON guide.
② Upstream GitHub
GitHub Repo ↗
Pinned to commit 63a65c1fcc.