event-prospecting

event-prospecting turns a conference or event speakers URL into a ranked list of people an account executive should talk to, each with a "why reach out" rationale. It solves event lead generation by extracting speakers from an event page, grouping them by company, filtering those companies against the user's ICP (ideal customer profile), then deep-researching only the speakers at ICP-fit companies. The final deliverable is a person-first HTML report where each card answers why the AE should talk to that person, with public links and a one-click DM opener, plus alternate company and people views and a results.csv for cold-outbound import.

The skill follows a fixed 10-step pipeline: setup output directory, load the user's ICP profile from profiles/{user_slug}.json, recon to detect the event platform, extract people, group by company, ICP triage with fast company-level scoring, filter by an icp_fit_score threshold (default 6), deep research on the fits, enrich speakers, and compile the HTML and CSV report. It enforces strict tooling: all web search goes through browse cloud search (never WebSearch), page extraction uses node scripts/extract_page.mjs (which falls back to browse get markdown for JS-heavy pages, never WebFetch), and research output is written as one markdown file per company or person via bash heredoc (never the Write tool). Subagents may use only the Bash tool, with hard tool-call caps (ICP triage 1 call/company, deep research 5 calls/company, person enrichment 4 calls/person). Strong anti-hallucination rules require product_description to quote a specific phrase from extract_page.mjs output or be marked Unknown, and forbid inferring anything from a site's fonts or design.

Target users are sales and GTM teams, particularly AEs prepping for events, building target lists from sponsor or speaker pages, and ranking prospects by ICP fit. It is invoked with an event URL and triggers on phrases like "find leads at {event}" or "research speakers at." Stated requirements are the browse CLI (npm install -g browse) and a BROWSERBASE_API_KEY environment variable; the same browse binary serves both API commands and JS-rendered page fallback. Output goes to a timestamped folder on the user's Desktop, and the skill requires at least one ICP profile to exist (example.json is only a template).