Imported from JeweAbiodun-sketch/research-agent (
AGENTS.md). Install upstream withnpx skills add JeweAbiodun-sketch/research-agent. Copyright stays with the author.
AGENTS.md
Autonomous Company Research & Report Generation Agent
Project: Module 3 — Individual Project
Framework: LangGraph + OpenAI + Pinecone + Cohere
Version: 1.0
Agent Name & Purpose
Name: CompanyResearchAgent
Purpose: Autonomously research any company and generate a structured 6-section investment due diligence report. Takes a single company name as input and delivers a complete report with no manual steps.
Tools Available
| Tool | Node | When to Use | API |
|---|---|---|---|
| OpenAI Web Search | research_node |
Always — for all 6 research topics | openai.responses.create with web_search_preview |
| OpenAI Embeddings | embed_and_store_node |
After research — to vectorise chunks | openai.embeddings.create (text-embedding-3-small) |
| Pinecone Upsert | embed_and_store_node |
After embedding — to store vectors | pinecone_index.upsert |
| Pinecone Query | retrieve_node |
After research is stored — to retrieve | pinecone_index.query |
| Cohere Rerank | rerank_node |
After Pinecone retrieval — to filter | cohere_client.rerank |
| OpenAI Chat | analyse_node |
After reranking — to classify | openai.chat.completions.create |
| OpenAI Chat | generate_report_node |
After analysis — to write sections | openai.chat.completions.create (gpt-4o) |
| Notion API | save_report_node |
After report generation — to save | POST /v1/pages |
Research Workflow (Step-by-Step)
1. VALIDATE → Clean company name, generate ticket ID (RES-YYYYMMDD-XXXXXXXX)
2. RESEARCH → Query OpenAI web search on 6 due diligence topics
3. EMBED → Chunk text (500 words, 50 overlap) → embed → upsert to Pinecone
4. RETRIEVE → Embed query → Pinecone top-10 similarity search
5. RERANK → Cohere cross-encoder → top-3 most relevant chunks
6. ANALYSE → LangGraph agent classifies risk, opportunity, priority
7. REPORT → GPT-4o generates 6 sections using reranked context
8. SAVE → Notion page created + local Markdown backup
Output Format — 6-Section Report
| # | Section | Content |
|---|---|---|
| 1 | Executive Summary | Overview, stage, sector, headline metrics |
| 2 | Business Model | Revenue streams, pricing, customer segments |
| 3 | Market Analysis | TAM/SAM/SOM, growth rate, regulatory environment |
| 4 | Competitive Landscape | Top 3–5 competitors, moat assessment |
| 5 | Risk Factors | Tech/market/team/regulatory/financial risks (Low/Medium/High) |
| 6 | Investment Recommendation | Score (1–10), Priority (Pass/Watch/Invest), next steps |
Error Behaviour
| Error Type | Behaviour |
|---|---|
| Empty company name | Pipeline stops at validate_input — returns error message |
| OpenAI API failure | 3 retries with exponential backoff (1s, 2s, 4s) |
| Pinecone upsert failure | 3 retries — if all fail, error logged and pipeline stops |
| Cohere rerank failure | Falls back to top-3 Pinecone similarity results |
| Report section failure | Section marked [Generation failed] — other sections continue |
| Notion save failure | Falls back to local Markdown file |
| Any node failure | error_handler_node fires — logs all errors, alerts analyst |
Skills Reference
| Skill File | Covers |
|---|---|
skills/research.md |
Web search prompts, topic coverage, chunking strategy |
skills/classification.md |
Risk scoring rubric, opportunity scoring, JSON schema |
skills/report_gen.md |
Section prompts, word limits, citation format |
Environment Variables Required
OPENAI_API_KEY = sk-...
PINECONE_API_KEY = pcsk_...
COHERE_API_KEY = ...
NOTION_TOKEN = secret_... # Optional
NOTION_DB_ID = ... # OptionalNotebook Structure
| Notebook | Sprint | Covers |
|---|---|---|
sprint1_foundation.ipynb |
Day 1 | AgentState schema, validation, acknowledgement |
sprint2_research.ipynb |
Day 2 | OpenAI web search, chunking, Pinecone storage |
sprint3_rag_agent.ipynb |
Day 3 | Pinecone retrieval, Cohere reranking, LangGraph analysis |
sprint4_5_report_pipeline.ipynb |
Day 4–5 | Report generation, full pipeline, live demo |