company-research

Company discovery and deep research skill. Researches a company's product and ICP, discovers target companies to sell to using Browserbase Search API, deeply researches each using a Plan→Research→Synthesize pattern, and scores ICP fit — compiled into a scored research report and CSV. Supports depth modes (quick/deep/deeper) for balancing scale vs intelligence. Use when the user wants to: (1) find companies to sell to, (2) research potential customers, (3) discover companies matching an ICP, (4)

3,619stars227forksUpdated 7/4/2026

Productivity & Workflow #browserbase #sales #lead-generation #market-research #automation

Security Assessment

Safe(100/100)

Security Score100/100

About company-research

The company-research skill discovers and deeply researches companies to sell to, aimed at go-to-market use cases such as finding prospects, building a target list, or doing market research against an ideal customer profile (ICP). It first develops a thorough understanding of the user's own company and product, then discovers candidate companies with the Browserbase Search API, researches each one using a Plan then Research then Synthesize pattern, scores ICP fit, and compiles the results into a scored research report and CSV. The pipeline is a fixed five steps -- company research, depth-mode selection, discovery, deep research and scoring, then report and CSV -- and depth modes (quick roughly 100 targets, deep roughly 50, deeper roughly 25) trade scale against research intensity.

Operationally it requires the browse CLI (installed via npm) and a BROWSERBASE_API_KEY environment variable. Tool usage is tightly constrained: web searches go through browse cloud search rather than WebSearch, page extraction uses the bundled extract_page.mjs script rather than WebFetch, and subagents write one markdown file per company via bash heredoc rather than the Write tool. All output lands in a dated folder on the user's Desktop containing per-company markdown files plus a final CSV, giving both a scored spreadsheet and the full research.

A strong set of anti-hallucination rules governs quality. Product descriptions must quote or paraphrase real page content from extract_page.mjs, cosmetic signals like fonts or frameworks may never be used to infer what a company sells, the user's own ICP must not leak into a target's description, and any company with no readable homepage content is marked Unknown with its ICP fit score capped.

FAQ

What does this skill require to run?

It requires the browse CLI (installed with npm install -g browse) and a BROWSERBASE_API_KEY environment variable.

What output does it produce?

It writes one markdown research file per company plus a final scored CSV into a dated folder on the user's Desktop, so the user gets both the spreadsheet and the full research files.

What are the depth modes?

Three modes balance scale against intelligence: quick (about 100 companies), deep (about 50), and deeper (about 25).

How does it avoid hallucinating company details?

Product descriptions must quote or paraphrase content returned by extract_page.mjs, it never infers products from fonts or frameworks, and if the homepage returns no readable content it writes Unknown and caps the ICP fit score at 3.

Which tools does it use for search and extraction?

All web searches use browse cloud search and all page extraction uses the extract_page.mjs script; it never uses WebSearch or WebFetch, and subagents are restricted to the Bash tool only.