| name | discover-company |
| description | Discover one noteworthy company (public or private) not yet in the knowledge base, then research it. Use when the user asks to "discover a company", "find me an interesting company", "surprise me with a company", "what company should I know about", or similar open-ended discovery requests. Picks a company based on recency, relevance to the user's domain (developer tools, AI infrastructure, web/cloud), and signal strength, then hands off to research-private-company or research-public-company. |
Discover a company
Goal: surface one company that is genuinely worth the user's attention right now — not already in the knowledge base — and produce a full research brief on it by invoking the appropriate downstream research skill.
Step 1 — Read the existing knowledge base
Check what's already tracked so you don't repeat it:
ls ~/knowledge/private-companies/ 2>/dev/null
ls ~/knowledge/public-companies/ 2>/dev/null
Build two sets: known_private (slugs) and known_public (tickers). A company is "already known" if its slug or ticker appears in either set.
Step 2 — Identify a candidate
Search broadly for companies that are making news right now. Run these in parallel:
AI developer tools startup 2026 fundinginfrastructure startup series a b 2026developer platform new launch 2026frontend web tooling company 2026AI agent framework company 2026database startup series funding 2026GitHub trending companies backednew startup product hunt launch 2026
Use the results to generate a candidate shortlist of 5–8 companies. For each candidate, mentally score it on:
- Novelty — not in
known_private or known_public
- Relevance — overlaps with: developer tools, AI/ML infrastructure, cloud/edge/CDN, frontend frameworks, databases, observability, security for web apps, or companies that compete with or complement Vercel
- Signal strength — recent funding, product launch, public traction, or notable team. Avoid companies that are just blog posts with no product.
- Recency — founded or made news in the last 18 months scores higher
Pick the single highest-scoring candidate that is not already known.
Step 3 — Determine public vs private
A company is public if it trades on a major exchange (NYSE, NASDAQ, LSE, TSX, etc.) under a ticker symbol. Check by searching <company> stock ticker if not obvious.
- Public → proceed to Step 4a
- Private → proceed to Step 4b
Step 4a — Research (public company)
Invoke the research-public-company skill with the resolved ticker. That skill handles all storage, git, and output.
Before invoking, announce to the user:
Discovered: () — . Researching now…
Step 4b — Research (private company)
Invoke the research-private-company skill with the company name. That skill handles all storage, git, and output.
Before invoking, announce to the user:
Discovered: — . Researching now…
Selection principles
- One company per run. Don't list multiple candidates to the user — pick the best one and go.
- Prefer private companies when the signal is equal — private companies are harder to track and benefit more from the KB investment.
- Avoid hype without substance. A company with a viral tweet but no product, customers, or funding is not worth a brief.
- Avoid obvious giants. Don't pick Stripe, GitHub, Cloudflare, or other household names unless something genuinely material just happened that warrants a first-time brief.
- Relevance to the user's world — the user works at Vercel. Weight companies in the developer platform, AI tooling, edge compute, CDN, frontend, or infrastructure-as-code space higher.
- If all candidates are already known, do a second broader search sweep before giving up. Only report "no new candidates found" if two sweep attempts both produce all-known results.
Output
The output is entirely produced by the downstream research skill (full brief in initial mode, changelog entry in delta mode). This skill adds only the one-line "Discovered" announcement before handing off.
Do not produce a separate summary or conclusion — the brief IS the output.
