Skip to content

README.en.md

Latest commit

 

History

History
169 lines (115 loc) · 7.83 KB

README.en.md

File metadata and controls

169 lines (115 loc) · 7.83 KB
Raw

report-helper · One-Sentence Deep Research Reports

AI Skill Deep Research PDF Report Codex Claude Code WeChat Official Account X GitHub

中文 README · Quick Start · Example PDF · Issues · Support

Author / Contact: 嘉然 Jiaran · WeChat official account: 嘉然学习笔记 · WeChat: evadebot · X: @_jiaran

report-helper is an AI Skill for long-form research report writing. Given a research topic, it guides an agent through source collection, source quality review, evidence organization, thesis formation, editorial review, and polished PDF generation.

From one sentence to a shareable PDF report: sourced, reasoned, and formatted.

It is designed for research on products, companies, people, concepts, value chains, policies, and trends. The goal is not to write a short summary, but to run a fuller research-writing workflow with traceable sources, explicit judgments, and a formal deliverable.

Quick Start

Send this to Codex, Claude Code, OpenClaw, or another skill-capable agent:

Install the report-helper skill from https://github.com/Jiaranbb/report-helper. After installation, remind me to restart or refresh the current agent as required; after restart, run python3 scripts/check_environment.py to check the environment.

If your environment supports the skills CLI:

npx skills add https://github.com/Jiaranbb/report-helper --skill report-helper

Codex users can also install manually:

python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
  --repo Jiaranbb/report-helper \
  --path . \
  --name report-helper

Restart or refresh the current agent as required after installation.

Highlights

  • 🧭 One-sentence start: provide a topic and focus; the agent runs the full report workflow.
  • 🔎 Research before writing: the workflow requires current data checks before drafting.
  • 🧾 Traceable sources: key factual claims are marked with citation IDs such as <sup>a1</sup>.
  • 📊 Qualitative + quantitative: company reports must include growth, earnings quality, cash flow, financing, valuation, or comparable-company analysis where data is available.
  • 🎨 Designed PDF output: background color, typography, spacing, headings, and footer styling are tuned for sharing.

Preview

The example below was generated with Codex GPT-5.5 standard speed in about 15 minutes. Quality and runtime depend on model capability, source availability, web-search quality, and topic complexity.

Run example:

report-helper demo

PDF preview:

report-helper PDF preview

Example report: 中国算力产业链深度分析报告.pdf

Example Prompts

Deeply research China's compute infrastructure value chain, focusing on 2026-2030 development trends, and run automatically.

Deeply research a company, focusing on business model, growth potential, valuation logic, and risk.

Write an AI compute industry development research report.

Suitable / Not Suitable

Suitable for

  • Long-form research on products, companies, people, concepts, value chains, policies, and trends
  • Reports requiring public sources, evidence chains, and source grading
  • Research tasks that need a formal PDF deliverable
  • Multi-agent "peer review" for important topics
  • Company and industry research that should combine narrative judgment with quantitative indicators

Not suitable for

  • Short definitions, quick summaries, or generic Q&A
  • Lightweight writing that does not need web research or source review
  • Final professional decisions in finance, law, medicine, or other regulated domains
  • Precise conclusions unsupported by public or user-provided data
  • Workflows that only want editable Markdown as the final deliverable

Workflow

report-helper follows a 7-step workflow:

  1. Scope alignment: confirm topic, type, motivation, focus, and execution mode.
  2. Source collection: prioritize current official, regulatory, filing, primary research, and authoritative media sources.
  3. Sufficiency audit: identify missing evidence; search again when gaps are severe.
  4. Judgment formation: organize key claims, evidence chains, and counterarguments.
  5. Drafting: select the right report structure; add quantitative analysis for company reports.
  6. Review loop: check facts, sources, structure, tone, and PDF delivery.
  7. Delivery: generate the PDF and keep intermediate notes and logs where configured.

First Run

Check the environment:

python3 scripts/check_environment.py

Install PDF rendering dependencies if needed:

python3 -m pip install markdown weasyprint

Copy config.example.json to config.local.json:

{
  "output_dir": "./output",
  "work_dir": "./output/work",
  "intermediate_dir": "./output/intermediate",
  "author": "Your Name or Organization"
}

Quality Notice

Report quality depends heavily on model capability. Codex GPT-5.5 is recommended for best results.

AI-generated reports can still be wrong, including drawing incorrect conclusions from correct facts. Treat outputs as learning references, and use "peer review" between multiple AI agents to challenge assumptions, ask hard questions, and verify reasoning.

Related projects

  • ecommerce-helper — a complete e-commerce asset-pack Skill from new-product research and RMB pricing to PDP and social content;
  • content-reader — agent skills for saving Xiaohongshu, Twitter/X, YouTube, and Bilibili content;
  • xhs-reader — save Xiaohongshu posts locally without logging in;
  • pdf-reader — convert PDFs into Markdown with page markers and quality metrics;
  • dreamy-photo — dreamy photo editing while preserving real subject details;
  • autoskin-codex — preview-first, reversible themes for the Codex desktop app;
  • jiucai-helper — a testable personal investment-decision Skill combining method and discipline.

See more original projects on Jiaranbb's GitHub profile.

About the author

Jiaran (Jiaranbb) — independent developer / AI Builder

I turn workflows I genuinely need into reusable AI tools and Skills.

License

MIT License. See LICENSE.