I Feel Lucky — One Stock Pick Per Run
Produce exactly ONE stock recommendation per run, selected by an objective, repeatable
algorithm — never by recalling past picks from memory. Every figure in the final report
must trace to a data result fetched during THIS run.
Language: respond in the language the user is using in the conversation.
Market scope: US equities ONLY by default. Widen to HK or CN only if the user
explicitly asked for that market in this invocation. CN picks are informational only
(no paper-trading support).
Step 1 — Account context (dynamic, never hardcoded)
Identify ONE target account in the scoped market — the account this pick would actually
be traded in. Selection order: (a) the account the user named or clearly implied in this
invocation; (b) otherwise the account whose mandate/playbook best matches the scoped
market and theme; (c) if still ambiguous, the most actively managed strategy account in
that market. If the runtime context already lists accounts, use it and skip list.
For that single account only, fetch summary, positions, and read_playbook —
as parallel calls in one response. Do NOT deep-fetch every account.
The report's portfolio section (Step 5/6) MUST name which account was used as the fit
reference, so a mis-targeted run is visible and the user can redirect ("run it against
account X" on the next invocation).
- If a target account exists: note current holdings, cash, and strategy style (growth /
value / GARP / balanced) — these drive candidate weighting (Step 3) and portfolio fit (Step 5).
- If no account
