hunter-research

Walking into a coffee chat or an interview knowing the business, the latest news and how the loop works is the cheapest edge there is. This skill builds that brief from sources anyone can check, and leaves clearly marked gaps for what only reading can fill.

What the user typed after /hunter-research: $ARGUMENTS

Tool in this folder (standard library Python):

python3 "${CLAUDE_SKILL_DIR}/brief.py" --help

What it reads and writes

• companies.json: for Google, Meta, NVIDIA, Netflix and Amazon, the SEC identity (CIK and ticker of the listed parent, for example Alphabet for Google), the careers site and the company's own RSS feeds. Any other company works with --ticker and --feed URL.
• loops.json: each company's interview loop. Notes marked official come from the company's own hiring and prep pages, each with its URL, read on 2026-10-15. Notes marked community-reported come from candidate reports and are labelled that way in every brief.
• brief.template.md: the brief's sections.
• Writes hunter/companies/<company>.md and never overwrites one that exists (it may hold the user's notes) unless asked with --force.

The rules

• Every fact has a source. A line in the brief either links a filing, a feed item or a company page, or it says "(fill in)". Fill those lines only from sources you open and read, and add the link.
• Numbers come from filings. Revenue, growth and headcount come from the 10-K, 10-Q or the 8-K results exhibit, quoted with the period. Never from memory.
• Official vs community. Keep the label on every interview note. If a recruiter's description differs, the recruiter wins.
• AI in interviews. Google, Meta and NVIDIA say AI tools are not allowed during interviews (NVIDIA: it disqualifies the candidacy). Prepare with Claude, never use it in the interview, and tell the user so.
• People. The brief covers the company and the team. Research on a person (the hiring manager) belongs to /hunter-connector and uses public professional information only.

Steps

1. SEC asks scripts to identify themselves. Ask the user once for a name and email for the User-Agent and save it as "sec_user_agent": "Name email" in hunter/profile.json (or pass --user-agent). Without it the brief skips filings and says so.

2. Build the brief:

   python3 "${CLAUDE_SKILL_DIR}/brief.py" build --company Google --role "Software Engineer" --team "Google Cloud, Infrastructure"
   python3 "${CLAUDE_SKILL_DIR}/brief.py" build --company "Acme Corp" --ticker ACME --feed https://acme.example/blog/rss

   Pieces on their own: brief.py loop --company Meta, brief.py sec --ticker NFLX, brief.py feeds --company NVIDIA --days 14.

3. Read the sources and fill the "(fill in)" lines: Item 1 of the 10-K in a few plain lines, what the news means for the team, the team's work from the posting and its public engineering posts, and the user's angle from their real resume.
4. Show the user the brief and the three things most worth knowing before a chat: what the team does, one recent event, and how the loop works.
5. Suggest /hunter-connector for the hiring manager and /hunter-interview for mock rounds built from the loop.

Example: the reel

Jordan Reyes (fictional) targets the Software Engineer role on Google Cloud, Infrastructure. brief.py build --company Google finds Alphabet Inc. (GOOGL, GOOG on Nasdaq) with headquarters on file with the SEC at 1600 Amphitheatre Parkway, Mountain View, CA; the 10-K for 2025 filed 2026-02-05; the Q2 10-Q filed 2026-07-23; and the 8-K of 2026-07-22 with item 2.02 (earnings results), whose exhibit is where revenue gets quoted from. It lists the latest Google Cloud Blog posts, and the loop: Google's own hiring page describes recruiter conversations, possible assessments such as a coding exercise, a panel of structured interviews scored on the same rubrics, no brainteasers and no AI tools, roughly 6 to 8 weeks; the coding, system design and "Googleyness and Leadership" split is labelled community-reported. Jordan fills in what the infrastructure team builds from the posting and the Cloud blog, then moves on to the connector.

What it never does

• Never states a fact without a source, and never fills a gap with a guess.
• Never researches individuals beyond public professional information, and never anyone's private life.
• Never fetches what a site asks scripts not to: SEC with an identifying User-Agent and a pause between calls, each company feed once, nothing behind a login.
• Never promises an interview outcome. A brief helps the user prepare; it does not predict an offer.