research

A Claude Code skill that builds deep, structured, up-to-date context on any company, person, or product from one or more URLs.

Designed for:

• Work trial prep ("I'm doing a trial for them tomorrow, what do they actually do?")
• Pre-meeting briefs ("I have a call with this startup, give me everything")
• Competitive analysis
• "I want to build something for them and need full context first"

What you get

Given a URL, produces:

research/<company-slug>/
  PRODUCT.md         what they do, who for, traction
  TECH.md            inferred stack, github activity, eng culture
  TEAM.md            founders, key people, advisors
  HIRING.md          open roles, what they look for
  COMPETITIVE.md     who they compete with, their angle
  WORK_TRIAL.md      (optional) what to build to impress them
  CONTEXT.md         800-1500 token digest, paste into any agent system prompt
  RAW/               source HTML / fetched markdown
  COMBINED.md        (optional, multi-URL) cross-entity analysis

Every research run is also auto-ingested into gbrain so future agents inherit the context permanently. Old runs accumulate — you build a personal company-knowledge layer over time.

Install

Requires Claude Code. Drop into your global skills dir:

git clone https://github.com/matthewkim323/research ~/.claude/skills/research

(No deps to install — pure LLM workflow using Claude Code's built-in WebFetch, WebSearch, and gbrain MCP.)

Use

In any Claude Code session (or via jabby if you have it):

research https://kalilabs.ai

Or multiple URLs:

research https://kalilabs.ai https://hermes.ai

For work-trial mode (adds a WORK_TRIAL.md with concrete build ideas):

research --mode trial https://example.com

Other flags: --name <slug> (override slug), --out <dir> (override output dir), --refresh (force re-fetch even if gbrain has recent pages).

gbrain integration

If gbrain is wired into your Claude Code as an MCP server, the skill will:

1. Check gbrain for prior research on the target before fetching (enrich, don't skip)
2. Always fetch fresh content (gbrain priors are background context, not a substitute)
3. Write each output file as a gbrain page (companies/<slug>/<section>)
4. Set last_researched frontmatter so staleness is visible to future queries

Query your accumulated research later:

gbrain query "what does kalilabs do"
gbrain query "who are kalilabs's competitors"
gbrain search "kalilabs tech stack"

License

MIT. Fork it, modify it, send PRs.

Credit

Inspired by stephenhungg/extract-site's skill structure. Built with Claude Code.