pronto-live-feed

ProntoNLP Live Feed

Generates a live Claude artifact showing the ProntoNLP home feed (or company-scoped feed): Top Movers by sentiment score change, Trending Topics, and recent/upcoming Earnings Call documents. Data is always fetched fresh when the artifact opens.

⛔ TOOL RESTRICTION: Never call showDocumentMindMap or deepResearch from this skill. Use only the tools listed in the steps below.

---

Step 0: Parse the Request

A. Detect context

User says	Context
"show me a feed", "open the feed", "live feed", "home feed", "show me the pronto feed", no company named	Home — market-wide, no company filter
"show me the [company/ticker] feed", "open [company] feed", "[company] live feed"	Company — scoped to that entity

B. Date ranges (all rolling — recomputed from today each run)

Range	Value
Top Movers current	sinceDay = today − 30 days
Top Movers prior	sinceDay = today − 90 days, untilDay = today − 30 days
Trends	timeframeDays = 90
Documents	Upcoming (future) + Recent (last 30 days)

---

Confirm Before Proceeding

After Step 0, before calling any tools, show a short summary and wait for user confirmation.

Show:

• Context: "Home Feed" or "[Company Name] Feed"
• Top Movers: Last 30 days vs prior 90 days · Earnings Calls · Ordered by sentiment score change
• Trends: Last 90 days · Earnings Calls
• Documents: Upcoming + Recent · Earnings Calls

Ask: "Ready to open the Live Feed. Reply yes to continue, or adjust anything above."

Do not call any tools until the user confirms.

---

Step 1: Resolve Company (Company Context Only)

Skip this step if context is Home.

Call getCompanies with the company name or ticker to resolve the canonical entity:

getCompanies(companyNameOrTicker: <name or ticker>)

Save:

• companyId — used for getTopMovers and getStockPrices
• companyName — used for getTrends, document fetches, artifact title, and header

If the company cannot be resolved, ask the user to clarify before continuing.

---

Step 2a: Parallel Batch 1

Fire all of the following simultaneously.

Top Movers:

getTopMovers(
  dateRange:  { gte: "<today − 30 days, YYYY-MM-DD>", lte: "now" }
  sortBy:     ["sentimentScoreChange"]
  limit:      10
  companiesIds: [<companyId>]   # include only for company context
)

Note: The MCP server auto-computes the prior period as sinceDay − 100 days to sinceDay and returns sentimentScoreChange directly. Do not make a separate prior-period call.

Trends:

getTrends(
  dateRange:     { gte: "now-90d/d", lte: "now" }
  documentTypes: ["Earnings Calls"]
  limit:         20
  sortBy:        "score"
  companyName:   <companyName>   # include only for company context
)

Note: getTrends does not accept companyId. Use companyName for company context.

Documents:

Home context — 2 calls in parallel:

# Recent (last 30 days, market-wide)
getDocuments(
  dateRange:              { gte: "<today − 30 days, YYYY-MM-DD>", lte: "now" }
  documentTypes:          ["Earnings Calls"]
  sortOrder:              "desc"
  excludeFutureDocuments: true
  size:                   50
)

# Upcoming (future, market-wide)
getDocuments(
  dateRange:              { gte: "<today + 1 day, YYYY-MM-DD>" }
  documentTypes:          ["Earnings Calls"]
  sortOrder:              "asc"
  excludeFutureDocuments: false
  size:                   30
)

After receiving getDocuments results, deduplicate by (companyName, documentDate) — keep one entry per document. Extract: companyName, date, document title. Limit to 30 recent and 20 upcoming.

Company context — 2 calls in parallel:

# Recent (current year/quarter)
getDocuments(
  companiesIds:           [companyId]
  documentTypes:          ["Earnings Calls"]
  excludeFutureDocuments: true
)

# Upcoming
getDocuments(
  companiesIds:           [companyId]
  documentTypes:          ["Earnings Calls"]
  excludeFutureDocuments: false
)

After receiving results, filter the upcoming call to keep only future-dated documents.

---

Step 2b: Parallel Batch 2

After Step 2a completes, use id and latestDocDate from each mover in the getTopMovers result.

sentimentScoreChange is returned directly by the tool — use it as-is. If absent for a mover, set to null.

Fire one getStockPrices call per mover, all simultaneously:

# Repeat for each mover (up to 10):
getStockPrices(
  companiesIds: [<mover.id>]
  dateRange: { gte: "<mover.latestDocDate − 8 days>", lte: "<mover.latestDocDate + 8 days>" }
)

Attach the returned price array to the mover object as stockPrices.

---

Step 3: Build the Payload and Delegate

Pre-format marketCap for each mover:

Raw value	Formatted
≥ 1 trillion	$1.2T
≥ 1 billion	$45B
≥ 1 million	$850M

Delegate to pronto-live-artifact agent (subagent_type: prontonlp-plugin:pronto-live-artifact). Do not render HTML here.

Pass the following structured payload:

artifact_type: live_feed
title: "ProntoNLP Live Feed"      # home context
      | "<companyName> Live Feed"  # company context

data:
  meta:
    context: "home" | "company"
    companyId?: <string>           # company context only
    companyName?: <string>         # company context only
    generatedAt: <ISO 8601 timestamp>
    topMovers:
      sinceDay: <YYYY-MM-DD>       # today − 30 days
    trends:
      timeframeDays: 90

  topMovers:                       # from current-period getTopMovers, enriched
    - id: string
      ticker: string
      name: string
      sector: string
      sentimentScore: number       # 0–1
      sentimentScoreChange: number | null  # from tool response; null if absent
      stockChange: number          # % market cap change
      marketCap: string            # pre-formatted
      latestDocDate: YYYY-MM-DD
      stockPrices:                 # from getStockPrices; omit if call failed
        - date: YYYY-MM-DD
          price: number

  trends:                          # from getTrends
    - name: string
      hits: number
      score: number
      change: number

  documents:
    upcoming: [ {companyName, title, date, documentType} ]  # future events
    recent:   [ {companyName, title, date, documentType} ]  # last 30 days

refresh:
  onOpen: true
  allowManualRefresh: true
  tools: [getTopMovers, getTrends, getDocuments, getCompanies, getStockPrices]
  params:
    context: "home" | "company"
    companyId?: <string>
    companyName?: <string>
    dateRangeMode: rolling

---

Step 4: Delivery

After the live artifact is ready, summarize in chat:

• Context: "Home Feed" or "[Company Name] Feed"
• Movers: N companies returned, sorted by sentiment score change
• Top mover: [Company name] with [+X.X% / −X.X%] sentiment score change
• Top trend: [Trend name], score [X.XX]
• Upcoming earnings: N documents scheduled
• Artifact: Lives in Claude and refreshes automatically on open

Do not mention tool names in the summary — describe results, not mechanics.

---

Best Practices

1. Never pass marketCaps or companyId to getTrends — use companiesIds for company context.
2. getTopMovers company filter uses companiesIds (array). Do not pass a separate prior-period call — the tool computes sentimentScoreChange internally.
3. Never fabricate — empty document buckets → pass empty arrays. Do not invent documents.
4. Stock prices are optional — if getStockPrices fails for a mover, omit stockPrices; the artifact skips the sparkline silently.
5. Do not mention tool names in user-facing messages — describe results, not API calls.