pronto-company-intelligence

Company Intelligence Report Generator

Produces a single-company intelligence report. Centerpiece: quarter-over-quarter comparison of every earnings call in the past year, explicitly showing whether sentiment, investment scores, and stock price reaction are RISING or FALLING. Layered on top: analyst forecasts, competitive benchmarks, trending topics, management quotes, and risk factors.

Data gathering and per-quarter analysis live here; final output is a regular standalone HTML report delegated to the pronto-html-renderer agent. This skill is not a live artifact.

⛔ TOOL RESTRICTION: Never call showDocumentMindMap or deepResearch. These are user-triggered only.

---

Step 0: Choose Report Mode

Mode	Use when	Tool calls	Sections included (payload keys)
Full Report (default)	Deep dive, no specific focus	~40	meta, kpi, quartersChart, stockChart, competitors, trends, speakers, quotes, predictions, risks
Quick Report	"quick", "overview", "brief"	~13	meta, kpi, stockChart, predictions
Sentiment Report	Earnings/sentiment focus	~25	meta, kpi, quartersChart, speakers, trends, quotes (bull/bear/forecast)
Competitive Report	Competitor/peer focus	~15	meta, kpi, stockChart, competitors
Risk Assessment	Downside focus	~20	meta, kpi, quartersChart, risks, quotes (bear/risk), speakers (analysts only)

Default to Full Report unless the user signals a narrower scope. See reference/report-template-guide.md for per-mode batch plans.

---

Confirm Before Proceeding

After Step 0, before calling any tools, present a short summary and wait for the user to confirm.

Show the user:

• Company: name and ticker as understood
• Report mode: which mode was selected and why
• Date range: e.g. "Past year — Apr 2025 to Apr 2026"
• Sections: brief list of what will be included

Then ask: "Ready to generate the report. Reply yes to continue, or clarify anything above."

Do not call any tools until the user confirms.

---

Tools Reference

#	Tool	Purpose	Company param
1	getCompanies	Company profile — companyId, sector, description	companyNameOrTicker
2	getCompanyPeers	Competitor/peer list	companiesIds
3	getDocuments	Earnings calls, 10-K, 10-Q — transcriptIds	companiesIds
4	getStockPrices	Historical OHLC price series	companiesIds
5	getStockChange	Price % change over a period	companiesIds
6	getCompanyConsensus	Analyst consensus estimates — all metrics in one call	companiesIds
7	getDocumentSummary (focus: 'key risks')	Risk factors from latest earnings call	transcriptsIds
8	getAnalytics	Sentiment scores, events, aspects	companiesIds
9	getTrends	Trending topics	companiesIds
10	getSpeakers (entityType: 'speaker')	Per-executive/analyst individual sentiment	companiesIds
11	getSpeakers (entityType: 'company')	Analyst firm aggregate sentiment	companiesIds
12	searchSentences	Key quotes from filings (via search-summarizer)	—

ID Flow

• getCompanies(companyNameOrTicker: ...) → companyId → pass to getStockPrices, getStockChange, getCompanyConsensus, getCompanyPeers
• getCompanyPeers(companiesIds: [companyId]) → peer companyIds → pass to getStockChange per competitor
• getDocuments(companiesIds: [companyId], ...) → transcriptId per doc → pass to getAnalytics, getDocumentSummary, and searchSentences per quarter

Always use companiesIds arrays (not companyName) when a tool accepts them.

---

Parallel Execution — Full Report

Batches run sequentially; within a batch, fire all calls simultaneously.

Batch 1 — foundation:

getCompanies(companyNameOrTicker: ...)    → save companyId, sector, description
getCompanyPeers(companiesIds: [companyId])  → save peer companyIds[]

Batch 2 — data collection (needs companyId):

getDocuments(companiesIds: [companyId], documentTypes: ['Earnings Calls'], excludeFutureDocuments: true)
                         → save transcriptId per earnings call
getStockPrices(companiesIds: [companyId], dateRange: {gte: 'now-1y/d', lte: 'now'}, interval: 'week')
getStockChange(companiesIds: [companyId]) ×3  (set dateRange per period: YTD / 6M / 1Y)
getCompanyConsensus(companiesIds: [companyId], metrics: ['revenue', 'epsGaap', 'ebitda', 'netIncomeGaap', 'freeCashFlow', 'capitalExpenditure'], timeframeInterval: 'quarter')
getTrends(companiesIds: [companyId], dateRange: {gte, lte})

Batch 3 — deep analysis (needs transcriptIds):

getAnalytics ×4          (one per earnings call, pass transcriptsIds: [transcriptId])
getStockPrices(companiesIds: [companyId]) ×4  (1 week window around each call date, interval: 'day')
getDocumentSummary(focus: 'key risks and risk factors mentioned by management', transcriptsIds: [doc_q1, doc_q2, doc_q3, doc_q4], corpus: ['S&P Transcripts'])
  → pass all available transcripts (max 5) for comprehensive multi-quarter risk view
getSpeakers(entityType: 'speaker', speakerTypes: ['Executives'], sortBy: 'count', companiesIds: [companyId])
getSpeakers(entityType: 'speaker', speakerTypes: ['Executives_CEO'], companiesIds: [companyId])
getSpeakers(entityType: 'speaker', speakerTypes: ['Executives_CFO'], companiesIds: [companyId])
getSpeakers(entityType: 'speaker', speakerTypes: ['Analysts'], sortBy: 'sentiment', sortOrder: 'desc', companiesIds: [companyId])
getSpeakers(entityType: 'company', speakerTypes: ['Analysts'], companiesIds: [companyId])
getStockChange(companiesIds: [peer1, peer2, peer3, ...], dateRange: {gte: 'now-1y/d', lte: 'now'})
  → ONE call with all peer IDs — not one call per peer

Batch 4 — quotes and forecasts (REQUIRED — do not render before this completes):

Delegate to ONE pronto-search-summarizer (subagent_type: prontonlp-plugin:pronto-search-summarizer):

Fetch all quotes needed for the [company] intelligence report. Run these searches:
1. Forecast/guidance quotes across all quarters — transcriptsIds: [doc_q1, doc_q2, doc_q3, doc_q4],
   topicSearchQuery: 'forecast guidance outlook', DLSentiment: ['positive'], size: 8
   → ONE call spanning all quarters (not 4 separate calls)
2. Most bullish executive quotes — companiesIds: [companyId], speakerTypes: Executives,
   DLSentiment: ['positive'], documentTypes: ["Earnings Calls"], size: 5
3. Top risk/bearish quotes — companiesIds: [companyId], topicSearchQuery: 'risk challenge headwind',
   DLSentiment: ['negative'], documentTypes: ["Earnings Calls"], size: 5
4. Notable analyst questions — companiesIds: [companyId], sections: EarningsCalls_Question,
   documentTypes: ["Earnings Calls"], size: 5
Return with speaker name, role, date. Citation link is already embedded in the text field.

Save the top 1–2 quotes per task, tagging each by section (bull, bear, forecast, risk).

Batch 5 — render (see Step 4 below).

---

Core: Per-Quarter Earnings Comparison

This is the heart of the report. Run getAnalytics separately for each of the past year's earnings calls — the only way to produce RISING/FALLING comparisons.

For each earnings call:

getAnalytics:
  companiesIds: ["<companyId>"]
  transcriptsIds: ["<transcriptId>"]
  documentTypes: ["Earnings Calls"]
  dateRange: { gte: "<quarter start>", lte: "<quarter end>" }
  analyticsType: ["scores", "eventTypes", "aspects", "patternSentiment", "importance"]
  → extract: sentimentScore, investmentScore, patternSentiment, top events, top aspects

getStockPrices (1 week window around call):
  companiesIds: ["<companyId>"]
  dateRange: { gte: "<7 days before call date>", lte: "<7 days after call date>" }
  interval: "day"
  → compute: % change before vs after = stock reaction

Compute for each quarter: sentimentDirection, investmentDirection, stockReaction. Compare first vs last quarter for overall trajectory.

Always surface in the narrative:

1. "Sentiment is [RISING/FALLING] — from X.XX (Q1) to X.XX (Q4)"
2. "Investment score is [RISING/FALLING] — from X.XX (Q1) to X.XX (Q4)"
3. "Stock reacted [positively/negatively] to [N] of [total] earnings calls"

Divergences are the most valuable insights. Flag them in the executive summary:

• Sentiment falling + stock rising → market disagrees with tone
• Sentiment rising + stock falling → market skeptical despite positive language
• Investment score falling while sentiment rising → deeper issues beneath the surface

---

Speaker Analysis

After Batch 3 completes, compute and include in the narrative:

• Average executive sentiment vs average analyst sentiment
• "Executives are MORE POSITIVE/MORE NEGATIVE than analysts by X.XX"
• "CEO is MORE BULLISH/MORE CAUTIOUS than CFO" (compare individual scores)
• Most bullish analyst: [Name] from [Firm] (X.XX); most bearish: [Name] from [Firm] (X.XX)
• Gap interpretation: > +0.10 = management may be over-optimistic; < -0.10 = street sees more upside

---

Step 4: Render

Delegate the HTML output to pronto-html-renderer (subagent_type: prontonlp-plugin:pronto-html-renderer). Do not render HTML here.

report_type: company
filename: <TICKER>-report-<YYYYMMDD>.html
title: "<Company Name> (<Ticker>) — Intelligence Report"
subtitle: "Generated: <date> · Sector: <sector> · Market Cap: <cap>"
data:
  meta: { ticker, companyId, companyName, sector, subSector, asOfDate }
  kpi:  { investmentScore, investmentScoreChange, sentimentScore, sentimentScoreChange,
          stockChangeYTD, stockChange6M, stockChange1Y }
  quartersChart:
    quarters:          [ "Q1 2025", "Q2 2025", ... ]
    sentimentScores:   [ ... ]
    investmentScores:  [ ... ]
    stockReactions:    [ ... ]
    positiveEvents:    [ ... ]
    negativeEvents:    [ ... ]
  quarterCards: [ { label, date, sentiment, sentimentArrow, investment, investmentArrow,
                    patternPos, patternNeg, revenue, badge, notes, isLatest } ]
  stockChart: { dates: [...], prices: [...], earningsCallIndices: [...] }
  competitors: [ { name, ticker, return1Y, isTarget } ]           # sorted desc, target first
  trends: [ { name, score, change, hits, explanation } ]
  speakers:
    executives: [ { name, role, sentiment, sentenceCount } ]       # incl. CEO, CFO, execAvg rows
    analysts:   [ { name, firm, sentiment, sentenceCount } ]
    gap:        { execAvg, analystAvg, interpretation }
  quotes: [ { text, speakerName, role, company, date, section } ]  # section ∈ {bull, bear, forecast, risk}; text ends with "[Source](url)"
  predictions: { revenue: [...], epsGaap: [...], ebitda: [...],
                 netIncomeGaap: [...], freeCashFlow: [...], capitalExpenditure: [...] }
  risks: [ { title, evidence } ]  # evidence contains "[Source](url)" if present
narrative:
  executiveSummary: "<2–3 paragraphs that explicitly state all RISING/FALLING verdicts, the exec-vs-analyst gap, and the thesis>"
  verdict: "<bullish / bearish / neutral + 3 supporting points>"

Report section order (renderer follows this sequence):

1. Header
2. Executive Summary
3. Stock Performance — stock chart + stock KPI tiles (YTD · 6M · 1Y)
4. Financial Outlook — predictions table (Revenue, EPS, EBITDA, Net Income, FCF, CapEx)
5. Quarter Cards — one card per earnings call
6. Competitors — peer return table
7. KPI grid (full — all 7 metrics)
8. Quarters chart (sentiment + investment lines + stock reaction bars)
9. Trends, Speakers, Quotes, Risks, Verdict

The renderer applies shared conventions (color rule, score display, company/citation links, chart palette, quarter-card styling, badge rules). For narrower report modes, omit payload keys that were not gathered — the renderer skips absent sections.

---

Step 5: Optional XLSX Export

After the HTML renderer reports success, ask the user:

"Your report is ready: <filename>.html. Want this also as an XLSX file? (yes/no)"

Skip the prompt if the user explicitly asked for XLSX up front (e.g. "give me the Tesla report as xlsx", "in spreadsheet form") — in that case generate both formats automatically.

If the user answers yes (or pre-asked), invoke anthropic-skills:xlsx directly from this skill (not via a sub-agent) using the same data you already built for the HTML renderer.

Filename: same as the HTML file but .xlsx extension.

Sheets to create (skip any whose source data is missing or empty):

1. Summary (tab teal #205262, no autofilter) — meta fields as Key / Value rows, then narrative.executiveSummary and narrative.verdict as wrapped text blocks
2. Stock Performance — stockChart dates × prices, followed by a separator and YTD / 6M / 1Y stock-change KPI rows
3. Financial Outlook — predictions flattened: one row per (Metric, Period, Estimate, Low, High, Actual) across all 6 metrics
4. Quarter Cards — one row per quarterCards entry: Quarter, Date, Sentiment, Investment, Revenue, Badge, Notes
5. Competitors — Company, Ticker, 1Y Return, Target (✓ for isTarget)
6. KPI Grid — all 7 KPI metrics as Label / Value rows
7. Quarters Chart Data — Quarter, Sentiment, Investment, Stock Reaction, Positive Events, Negative Events
8. Trends — Topic, Score, Change, Hits, Explanation
9. Speakers — Type (Executive / Analyst), Name, Role / Firm, Sentiment, Sentences; gap footer rows below
10. Quotes — Section, Quote (with embedded Source link), Speaker, Role, Company, Date
11. Risks (tab red #ED4545) — Risk, Evidence (with embedded Source link)

Styling (every sheet):

• Row 1: fill #205262, white bold text, height 22pt, frozen so it stays visible when scrolling
• Autofilter on header row (all sheets except Summary)
• Positive numeric values → font #6AA64A (green) · Negative → #ED4545 (red)
• Scores: 0.00 · Change/% columns: 0.0% · Counts: whole numbers
• Hyperlinks: blue underlined, display text "Source"
• Wrap long text (quotes, narratives) — no column wider than ~50 chars
• No zebra striping · No cell borders

Report the saved filename to the user when complete.

If the user answers no, end the skill normally.

---

Date Handling

Past year:     dateRange: { gte: "now-1y/d",  lte: "now" }
Past quarter:  dateRange: { gte: "now-90d/d", lte: "now" }
YTD:           dateRange: { gte: "<YYYY>-01-01", lte: "now" }
Past 6 months: dateRange: { gte: "now-6M/d",  lte: "now" }

Always pass dateRange with explicit gte and lte to every tool call that accepts it.

getAnalytics max range: 1 year — split longer requests into multiple yearly calls.

---

Error Handling

Problem	What to do
Company not found	Try ticker instead of name (or vice versa); check spelling
getCompanies returns no match	Ask user to verify; do not proceed without companyId
Fewer than 4 earnings calls	Work with available quarters; note the gap in the payload
No predictions for a metric	Omit from payload; renderer skips absent rows
Analytics returns empty	Verify date range ≤ 1 year; try without transcriptsIds filter
No competitors returned	Omit competitors from payload
No quotes returned	Omit quotes — never fabricate
Private/unlisted company	ProntoNLP covers public companies only — tell the user
companyId missing from response	Inspect the full object for id or nested field

---

Best Practices

1. Save companyId the moment you get it from getCompanies.
2. Maximize parallelism — batch all independent calls per the strategy above.
3. Never fabricate data — if a tool returns nothing, omit the corresponding payload key.
4. Never reformat raw scores — pass 0.71, not 7.1 or 7.1/10 (renderer enforces).
5. Always pass companiesIds: [companyId] (array) — never pass a bare companyId string to tools.
6. Do not mention tool names in responses — describe the action ("I analyzed 4 earnings calls", not "I called getAnalytics 4 times").