idea-generation · openai — Claude skill: triage public-equity ideas

What idea-generation does

This skill handles the first pass over public-equity candidates. Feed it screens, notes, files or connected sources, and it returns candidates grouped by archetype: long, short, pair, relative value, catalyst, watchlist. False positives get marked as rejected instead of quietly promoted.

The workflow is spelled out step by step. Set the mandate and universe, normalize tickers, ADRs, share classes, currencies and fiscal calendars, map thematic names by beneficiary pathway, then run PM triage on denominator quality, estimate credibility, catalyst path, liquidity and false-positive risk. For structured input there is scripts/score_ideas.py, which takes a CSV of candidate rows and scoring fields and writes ranked support output plus warnings. It does not produce market data of its own.

One rule is explicit: a screen result is never a final recommendation. This is a research-prioritization tool, not a trade advisor. The crowded label is reserved for cases backed by positioning, ownership, flows or short interest; when the only support is price appreciation or narrative visibility, the output uses expectations-heavy, valuation-gated or crowding-risk candidate. A name does not advance just because it rallied after a thematic catalyst — the driver has to link to orders, backlog, revenue, margins or estimate revisions, or it gets tagged needs exposure attribution.