company-investigation

Overview

This skill performs a structured, multi-source investigation of a company before a meeting, sales call, interview, or due diligence review. Given a company URL, it fetches the company's website, dispatches four parallel research agents, then synthesises all findings into a single self-contained HTML report saved to the current working directory.

The report covers 10 sections, tags every claim with a source credibility badge, surfaces an AI Moat Score, and delivers ready-to-use talking points.

---

Tools Used

• WebFetch — read the company URL and its subpages
• WebSearch — research competitors, news, funding, tech stack, and sentiment
• Agent — dispatch four parallel research agents
• Write — save agent findings to /tmp/ and write the final HTML report
• Read — read reference files and agent output files

---

Step 1: Quick Scan

When the user invokes /investigate <url>:

1. Use WebFetch to read the company URL provided by the user.
2. Extract: company name, tagline, what they do (1–2 sentences), key subpages found in navigation.
3. Present a brief summary (3–5 sentences: who they are, what they build, who they serve).
4. List the 10 report sections that will be covered:
   1. Company Overview
   2. Product/Service Analysis
   3. Team/Leadership
   4. Business Model
   5. Market & Competitors
   6. Financials & Funding
   7. Tech Stack / Engineering
   8. News & Sentiment
   9. Red Flags / Risks
   10. AI Moat Analysis
5. Ask: "Want to adjust focus areas, add context, or skip any sections before I run the full investigation? (Or just say 'go' to proceed with all sections.)"
6. STOP. Wait for the user's next message before doing anything else.

Step 1 error handling

• URL unreachable: Inform user, offer web-search-only mode, note in report header.
• Empty/minimal content: Note to user, use web search as primary source, continue.
• Malformed URL: Ask user for a valid URL before continuing.

---

Step 2: Deep Research

After the user confirms, note {{company_name}} from Step 1. Read references/agent-prompts.md for the full instructions to pass to each agent. Then dispatch 4 parallel agents simultaneously — do not wait for one before starting the next.

Agent	Sections	Output file
Company Agent	1–4	/tmp/investigation_company_agent.md
Market Agent	5	/tmp/investigation_market_agent.md
Financial/News Agent	6–8	/tmp/investigation_financial_news_agent.md
AI Moat Agent	9–10	/tmp/investigation_ai_moat_agent.md

Wait for all four agents to complete before proceeding to Step 3.

---

Step 3: Synthesis & Report Generation

3.1 — Read all agent output files

Read each file in full. If a file is missing or empty, note which sections are affected — mark them "Research unavailable for this section" in the final report.

3.2 — Resolve conflicts

Compare findings for factual contradictions. For each conflict: retain both claims, prefix with CONFLICT: in bold, show both claims with sources.

3.3 — Source validation pass

Verify every claim carries a credibility tag ([Verified], [Reported], or [Unconfirmed]). Any claim missing a tag → assign [Unconfirmed].

3.4 — Generate the HTML report

Read references/html-template-guide.md for complete HTML generation rules before writing. Read templates/report.html for the structural layout and placeholder conventions.

Generate a complete, self-contained HTML report replacing all {{placeholder}} variables with actual content. Include all 10 sections, Executive Summary, AI Moat Score widget, Talking Points, and Source Overview.

3.5 — Save the report

Filename pattern: {company_name_lowercase_hyphenated}-investigation-{YYYY-MM-DD}.html

Write to the current working directory (not /tmp/).

3.6 — Confirm completion

Tell the user: "Investigation complete. Report saved as {filename} in the current directory." Provide the full absolute file path.

3.7 — Open in browser (macOS)

Offer to open the file automatically: open {filename}

---

Reference Files

File	Purpose
references/agent-prompts.md	Full instructions to pass to each of the 4 research agents
references/source-credibility.md	Three-tier credibility framework and rules for tagging claims
references/report-sections.md	Definitions for all 10 sections, key questions, and output formats
references/html-template-guide.md	HTML generation rules, no-JS constraint, placeholder mapping
templates/report.html	Structural HTML template with {{placeholder}} variables and CSS

---

Output Spec

Property	Value
Filename	{company_name_lowercase_hyphenated}-investigation-{YYYY-MM-DD}.html
Location	Current working directory
Format	Self-contained HTML (no external dependencies)
Encoding	UTF-8
Auto-open	Offered after save on macOS; file path always shown

---

Error Handling

Condition	Response
URL unreachable at Step 1	Offer web-search-only mode; note in report header
URL returns empty/minimal content	Use web search as primary; continue normally
Company too obscure	Deliver what is available; flag low-coverage sections [Unconfirmed]; add notice at report top
Agent fails or times out	Mark affected sections "Research unavailable"; note in report header
Rate limiting during agent runs	Degrade gracefully; note affected sections; do not halt investigation
Conflicting facts between agents	Apply CONFLICT resolution from Step 3.2; never silently discard a claim
Template file missing	Use fallback inline HTML structure; warn user template was not found