Skip to main content
Deep Research & Company Intelligence Skill

Agent skill · 3 min read

Deep Research & Company Intelligence Skill

An AI agent skill for Claude and Codex that turns a company name or website into a current, cited intelligence report covering ownership, leadership, financials, technology, competitors, opportunities and risks.

FreeClaude Code · Claude apps · Codex

Free preview

A company report written by an AI agent is easy to produce and hard to trust. Two companies with similar names get merged. A two-year-old headcount is presented as current. A partner logo becomes a customer. And nothing in the report tells you which sentences are facts.

Deep Research & Company Intelligence gives your agent the discipline of a corporate-intelligence analyst: resolve the entity, go to primary sources, date everything, and say how sure you are.

The trouble with summarising search results

Search results are for discovery, not evidence. Without rules, an agent:

  • trusts the first aggregator it finds
  • silently picks one number when sources disagree
  • confuses a registered office with headquarters, or a job post with a live project
  • fills gaps with plausible-sounding detail

How the skill works

  1. Define the objective — who the report is for, which decision it supports, and 5–15 research questions before browsing.
  2. Resolve the entity — legal name, registration numbers, domains, parent and subsidiaries. No synthesis until identity is settled.
  3. Rank the sources — registries and filings first, aggregators last, each source scored for authority, directness, freshness and independence.
  4. Collect evidence across eighteen categories, including ownership, leadership, business model, customers, locations, financials, funding, technology, hiring, reputation and legal risk.
  5. Verify and date — critical claims are checked against two strong sources wherever practical, every time-sensitive fact carries a date, and contradictions are reported rather than hidden.
  6. Red-team the report — search for evidence against its own conclusions and re-check the latest news before finishing.

Every claim is labelled

The report separates verified, reported, estimated, inferred and unknown. Here is how the skill tells your agent to handle a contradiction:

Employee count is uncertain. LinkedIn lists 501–1,000 employees, while a 2025 press release states 1,200 staff globally. The difference likely reflects timing or group scope.

Choose the depth and the format

Seven modes — company profile, sales intelligence, competitive intelligence, vendor evaluation, due-diligence support, market, and trigger monitoring — at four depths, from a 5–10 source quick profile to maximum-coverage due-diligence support. Output as a full cited report, a one-page intelligence card, or JSON.

Built for international research

It searches local-language and transliterated names and legal-suffix variants, uses national company registries, and keeps group headquarters, country subsidiaries, branches and special-jurisdiction entities apart.

Who it's for

Sales teams researching accounts, founders sizing up partners or competitors, and analysts who need a first pass they can defend.

What it won't do

It won't bypass paywalls or logins, dig into employees' private lives, or present open-source research as legal, financial or investment advice.

The rest ships in the skill

That was the brief. The full playbook your agent follows — every phase, rule and output format — is in the download.

What your agent gets

  • Resolves the correct legal entity before any analysis
  • Every material claim is dated, cited and given a confidence level
  • Checks its own conclusions and runs a freshness check before delivering
  • Seven modes: profile, sales, competitive, vendor, due diligence, market and monitoring
  • International research: local-language names, legal-suffix variants and national registries
  • Output as a full report, a one-page card, or JSON
What's in the package12 files
  • SKILL.mdPrinciples, research modes, entity resolution, stop conditions, anti-hallucination rules, ethics
  • references/sources-and-discovery.mdSource hierarchy and scoring, discovery queries, source-specific and international guidance
  • references/evidence-collection.mdThe eighteen evidence categories, from ownership to legal risk
  • references/verification.mdContradictions, confidence labels, provenance, red-team review, freshness, citations
  • references/analysis.mdStrategic analysis, SWOT, competitors, positioning, opportunities, triggers
  • references/output-formats.mdReport template, intelligence card and JSON schema
  • references/source-register.mdDated data-provider facts, with sources
  • examples/entity-ambiguity-brief.mdA finished brief on a fictional company whose legal entity could not be fully resolved
  • assets/templates/research-plan.yamlResearch plan: decision, questions, evidence thresholds, budget, stop conditions
  • README.mdInstall steps for Claude Code, Claude apps and Codex, requirements and example requests
  • CHANGELOG.mdWhat changed in each version
  • LICENSE.txtLicence terms: free to use within your organisation
Free

Enter your email and the ZIP downloads straight away; we send you a copy of the link too.

Free to use and adapt within your organisation. Please don't republish or resell the files.

Details

Works with
Claude Code, Claude apps, OpenAI Codex, any agent that supports SKILL.md
Requires
Web search and page fetching
Format
SKILL.md + 6 reference files, a worked example and a starter template (ZIP)
Version
1.1.2
Price
Free, with your email
Licence
Use within your organisation; no public redistribution

Questions

Which agents does it work with?

Claude Code, the Claude apps (web and desktop) and OpenAI Codex, plus any agent that supports the SKILL.md format. The zip includes install steps for each.

What does my agent need?

Web search and page fetching. Paid databases such as Crunchbase or PitchBook are optional; the skill treats them as supporting evidence and still checks material facts against primary sources.

What exactly do I receive?

A ZIP containing the skill folder: SKILL.md, its reference files, a worked example, a starter template, a README with install steps, a changelog and the licence. It downloads as soon as you enter your email, and we email you a copy of the link.

Is it really free?

Yes. There is no payment and no card. We ask for your email to send you the download link, and we may follow up about the skill. If the download does not work, email support@octivex.tech.

Can my team use it?

Yes. Use and adapt it anywhere in your organisation, including private internal repositories and team skill libraries. Please don't resell it or republish it publicly.

More skills

See all skills