Stock Research Publish

Take the analysis outputs produced by stock-research-pipeline and turn them into a deliverable the user can read or share. Two output modes — PDF (print-style) or HTML (web-style with optional Netlify deploy). The user picks.

Prerequisites

• An NLM analysis already exists for the SYMBOL — i.e. the 6 query answers from stock-research-pipeline are available in this session OR can be re-derived from data/companies/{SYMBOL}/. If neither is true, run /stock-research-pipeline {SYMBOL} first.
• For PDF: Python venv at .venv/ with reportlab installed.
• For HTML deploy: netlify CLI on PATH OR a Netlify access token in $NETLIFY_AUTH_TOKEN. If the CLI is missing, the skill installs it (with user confirmation) — see Step 4 pre-flight. Requires npm on PATH (Node ≥ 18).

Step 1: Ask the user what they want

Use AskUserQuestion with one question, three options:

• PDF only — print-style A4 report (existing reportlab generator).
• HTML only — single-file, magazine-quality web page with optional Netlify deploy.
• Both — produce PDF first, then HTML, then ask about deploy.

Also ask whether to deploy the HTML to Netlify if HTML is selected (yes / no). Do not branch on whether netlify is installed at this stage — the Step 4 pre-flight handles install + auth + site name + deploy scope interactively.

Do not proceed until you have explicit answers. No defaults.

Step 2: Locate the analysis inputs

Resolve the SYMBOL from the user's request and gather the 6 NLM answers. Sources, in order:

1. The current conversation (if /stock-research-pipeline just ran in this session).
2. data/companies/{SYMBOL}/analysis/ — look for any of: business_model.md, industry.md, management.md, financials.md, growth_triggers.md, risks_scenarios.md, or a single nlm_answers.json if the pipeline saved one.
3. As a last resort, re-run only the missing NLM queries from the pipeline (do NOT re-download PDFs or recreate the notebook). Use the existing notebook ID from data/companies/{SYMBOL}/analysis/notebook.txt if present.

If you can't gather all 6 answers, stop and ask the user to run /stock-research-pipeline first.

Also load data/companies/{SYMBOL}/screener_data.md for headline metrics (PE, ROE, ROCE, market cap, latest revenue/PAT) — these populate the KPI cards.

Step 3a: PDF path

Use the existing reportlab generator at scripts/generators/equity_report_pdf.py (or, if absent, the working reference at data/companies/GRAVITA/analysis/generate_report.py). Render to:

data/companies/{SYMBOL}/analysis/{SYMBOL}_Equity_Analysis_{YYYY-MM-DD}.pdf

Then open the file and tell the user the path. Done if PDF-only was chosen.

Step 3b: HTML path — design rules

Generate a single self-contained .html file (no external assets except Google Fonts and one optional Mermaid CDN). Apply the visual-explainer playbook — these are not suggestions.

Aesthetic — pick ONE per generation. Vary across runs.

• Editorial (default for equity reports): Instrument Serif or Crimson Pro headlines, IBM Plex Sans body, JetBrains Mono for monospace. Generous whitespace, deep navy + muted gold, serif drop-caps on the executive summary.
• Blueprint: subtle grid background, deep slate/blue, monospace labels, precise borders. Good for technical/cyclical companies.
• Paper/ink: warm cream #faf7f5 background, terracotta + sage accents. Good for consumer / FMCG names where warmth fits.

Forbidden — do not produce these.

• Inter / Roboto / system-ui as --font-body. Pick a real pairing.
• Indigo/violet accents (#8b5cf6, #7c3aed, #a78bfa), cyan-magenta-pink neon, gradient text on headings.
• Emoji icons in section headers. Use styled monospace labels with colored dot indicators or numbered badges.
• "Neon dashboard" or gradient-mesh backgrounds. Use subtle radial glows or faint grid only.
• Animated glowing box-shadows or pulsing effects on static content.
• Three-dot window chrome on code/quote blocks.

Required structure (mirror the PDF section list, but use web idioms):

1. Hero — Company name (serif, 48–64px), SYMBOL in monospace, sector tag, report date. One-line investment thesis as a lead paragraph (larger, dimmer body color).
2. KPI strip — 4–6 cards: CMP, MCap, P/E, ROE, ROCE, latest QoQ growth. Hero number large; label small and dim. Vary one card as the "primary" with accent-tinted background.
3. Executive summary — 2–3 short paragraphs synthesizing the thesis. Optional pull quote for the single most important management line.
4. Business model & value chain (Query 1) — CSS Grid cards, not prose dump. Extract the revenue streams, unit economics, key cost drivers. If there's a flow (input → processing → output → customer), use a small Mermaid graph LR (≤6 nodes) with the full diagram-shell zoom container — never bare <pre class="mermaid">.
5. Industry & competitive positioning (Query 2) — 2-column layout: structure cards on left, peer comparison <table> on right (sticky header, alternating rows, right-aligned numbers with tabular-nums).
6. Financial deep dive (Query 4) — A real <table> for the quarterly P&L. Add a Chart.js line chart for revenue + margin trajectory if 4+ quarters available. Margin trend gets its own KPI callout.
7. Management quality & execution (Query 3) — Guidance-vs-delivery as a striped table. Italic gray management quotes inset with a subtle left-accent bar (no quote-mark emoji).
8. Growth triggers (Query 5) — Card grid, each card = one trigger with: name, evidence, sizing, kill-switch. Color the left border by conviction (high / medium / watch).
9. Variant perception scorecard (Query 5) — This is a flagship section, not a quick table. It must include: (a) A 1–2 sentence methodology note explaining what "consensus" means here (sell-side target, peer trading multiple, common buy-side view) and what the edge is being measured against. (b) A wide <table> with at least these columns: factor | consensus view (with a sourceable anchor — sell-side target, peer multiple, prior management commentary) | variant view (the differentiated take) | quantitative evidence (specific numbers, dates, data points that support variant) | edge size (in EPS / multiple / target terms — i.e. how much this factor moves the price if right) | time horizon (when this should resolve) | falsification (what would prove the variant wrong). (c) A minimum of 5 factors, not 3. Cover at least: valuation multiple, the most-feared risk (commodity / cyclical / regulatory), execution / capex, demand durability, and one sector-specific factor. (d) Below the table, a short "consensus build vs variant build" comparison — two side-by-side mini-tables showing the EPS/revenue/multiple stack each view implies for the next 2 years.
10. Bull / Base / Bear scenarios (Query 6) — Two complementary views, both required: (a) Three side-by-side scenario cards with FY+1/+2/+3 headline numbers, probability weight, return %, and one-line trigger. Bear uses rose/cranberry, base neutral, bull sage/emerald — border color only, never red/green emoji or arrows. (b) A detailed scenario <table> below the cards, with scenarios as columns and rows covering at minimum: Revenue FY+1/+2/+3, Revenue CAGR, EBITDA FY+3, EBITDA margin %, PAT FY+3, EPS FY+3, EPS CAGR, exit P/E multiple, target price, implied return %, and probability. Numbers are right-aligned with tabular-nums. Header row tinted to match the scenario card colors so the table is visually anchored to the cards above. (c) A scenario assumptions sub-table listing the 4–6 key driver assumptions per scenario (e.g. for a recycler: spread per kg, capacity utilization %, organized share %, non-core revenue mix, tax rate). This makes the s