Banker Memo

Banker-grade research memo from raw-data/ snapshots — prompt-driven, not template-driven.

What this skill does

Takes a deliverable dir that already has raw-data/*.json (produced by the CN Phase 3.5 pipeline) and drives an agent through a senior sell-side analyst framework to produce:

1. analysis.md — 8-section banker research memo (2500-4000 words, peer-benchmarked, with explicit data flags)
2. slides-outline.md — PPT blueprint (10-15 slides, content-driven — NOT a fixed 8-slide template)
3. data-provenance.md — every hard number traceable to a raw-data filename stem

Why prompt-driven (not Python template)

The 0.9.4-era build_deck.py used an 8-slide Python template that stamped the same structure onto every company. Problem: the result is a data dashboard, not banker analysis — no industry context, no peer benchmarking, no SOTP reasoning, no "4C's" credit framework, no banker opinion.

Prompt-driven flips this: the agent (which has read thousands of real sell-side reports during training) writes the narrative; the Python layer only does orchestration (dispatch + gate audit).

Best-fit cases