GP Report Builder

Generates a fully styled, self-contained HTML investor report for a named portfolio company. Supports two data source modes, two report types, and automatic brand detection from uploaded templates. Read this skill fully before writing a single line of HTML.

Requirements: A connected Chronograph MCP server. These workflows are designed for permissioned Chronograph users to connect to their private investment data, or an uploaded Excel model for Model mode.

Step 0 — Resolve Branding

Brand tokens are resolved automatically — the user never fills in a config file manually. Execute the following decision tree before any other step.

0A — Check for an Uploaded Brand Template

Look for any of the following in the current conversation:

Once extracted, map findings to the Brand Token Schema in 0C.

If the user has dropped multiple files, treat the most recent one as authoritative for branding, unless the user specifies otherwise.

0B — No Template Provided → Apply Chronograph Defaults

If no brand template is present in the conversation, apply the following defaults silently — do not ask the user to provide branding.

firm_name:             Chronograph
website:               www.chronograph.pe
confidentiality_label: CONFIDENTIAL

colors:
  bg_primary:       #101C1D   ← Near Black 1 (page background)
  bg_secondary:     #1A2627   ← Near Black 2 (card / panel background)
  bg_header:        #1B4147   ← Deep Teal (header strip)
  accent_primary:   #57E5EE   ← Bright Teal (headlines, KPI values, eyebrows)
  accent_secondary: #11A8B2   ← Regular Teal (bars, borders, left-rule accents)
  accent_negative:  #F95532   ← Accent Red (EBITDA bars, negative deltas)
  accent_positive:  #4ecb8a   ← Green (positive deltas)
  text_primary:     #FFFFFF   ← White (body text)
  text_muted:       #D9E8E8   ← Light teal tint (footnotes, commentary)
  table_header_bg:  #1B4147   ← Deep Teal (table header row)

fonts:
  heading:         Ubuntu
  heading_weight:  700
  body:            Open Sans
  body_weight:     300
  google_fonts_url: https://fonts.googleapis.com/css2?family=Ubuntu:wght@700&family=Open+Sans:wght@300;300i&display=swap

logo:
  url:      (none — render firm name as text)
  position: top-left

theme: dark

0C — Brand Token Schema

Whether tokens are extracted from a template (0A) or defaults are applied (0B), resolve all of the following before proceeding. Every downstream panel references these names.

0D — Theme Handling

Dark theme (luminance of bg_primary < 0.2): Use the token values as resolved. Default contrast pairings apply (white text on dark backgrounds). Minimum contrast ratio: 4.5:1 for body text, 3:1 for large text (≥ 18px bold).

Light theme (luminance of bg_primary ≥ 0.2):

• Swap text_primary to a near-black (e.g. #1A1A1A) if not already dark
• Ensure accent_primary provides ≥ 3:1 contrast against bg_primary
• Table header: use the firm's dark brand color (e.g. deep navy / dark green) rather than a light value

0E — Confirm with the User (optional, brief)

After resolving tokens, output a single short line — not a table, not a list — before generating the report:

"Using [firm_name] branding — [accent_primary] accent on [bg_primary] background, [font_heading] / [font_body] fonts."

If brand detection produced low-confidence results (e.g. only a logo image was provided with no color context), ask one focused question:

"I've picked up [X] and [Y] from your file — is that the right color scheme, or would you like to adjust anything?"

Do not ask if defaults were applied — just proceed.

Step 1 — Determine Report Mode

Data source mode:

Report type:

If unclear, default to GP One-Pager.

Step 2 — Resolve the Company & Fetch Data

MCP Mode

All data is fetched from the user's connected Chronograph MCP server. The agent must have the Chronograph MCP connected before running the skill in MCP mode; if it isn't connected, prompt the user to connect it, or fall back to Model mode if a file is available.

Inspect the connected Chronograph MCP's tool list at runtime and pick the appropriate tool for each step below based on the tool descriptions the server provides. Do not hard-code tool names — read the live descriptions to stay current.

Fetch sequence

Work through these steps in order. At each step, identify the right Chronograph tool from its description, call it to fetch what the report needs, and hold the result for the rendering steps that follow. The skill does not need a fixed schema — only the values listed below in plain language.

1. Resolve the company. Find the tool that searches for companies, funds, and other portfolio entities by name. Use it to turn the user's input into a canonical company. If multiple candidates come back, prefer the closest name match; ask the user only if genuinely ambiguous.

2. Get the company's basic facts. Use the tool that retrieves core portfolio entity records. Fetch what the Header Strip (Step 4) needs: company name, sector, industry, geography, HQ, and the company's reporting currency. Request only what the header renders — the tool's description will list what's available.

3. Get the company's investments. Using the same core-entity retrieval tool, fetch the list of investments associated with the company. The Investment Returns Table (Step 4) needs one row per investment: investment name, the fund it belongs to, entry date, exit date if applicable, and whether it has exited.

4. Get the company's financial line items. Find the tool for company-level financial metrics. Make a discovery / help call first, passing the company ID, to see what's available.

Metric selection rule. Always query by the platform's metric type when one is available for the line item — even if the company's display label differs. Use a company-specific mapped metric definition only when no metric type covers the line item or the request is inherently tenant-specific (bespoke KPIs, commentary fields, operating metrics, custom valuation inputs).

Priority:

1. Platform metric type
2. Company-specific mapped metric definition (only if no metric type exists)
3. Name-based metric search (only if neither of the above resolves)

Do not pick a company-specific metric because its label is a closer word match than an available metric type. Metric types drive querying; display labels drive presentation.

What the report needs (query via the metric type whenever available):

• Financial Performance Table: Revenue, Gross Profit, EBITDA, the company's adjusted EBITDA series, and Net Debt. Pull the last 4–5 trailing-twelve-month periods plus the most recent quarter.
• Valuation Summary Table: Enterprise Value, valuation multiple, Total Debt, Cash, Net Debt, Equity Value — current quarter and prior quarter.
• KPI Strip: the latest values for LTM Revenue, LTM Adj. EBITDA, Enterprise Value, and Net Debt.

Use LTM for income-statement items, As-of for balance items. The discovery response will tell you which period types are supported per metric.

5. Get per-investment returns. Find the tool for investment-level performance. It uses gross performance figures (not net — net lives elsewhere on the platform and is the wrong tool for the GP one-pager). The Investment Returns Table needs, per investment: invested capital, realized proceeds, unrealized value, MOIC, and IRR. Sum across investments to populate the MOIC card in the KPI Strip.

6. Resolve any non-standard metrics by name. If the user asks the report to include a tenant-specific KPI that isn't part of the platform's standard metric set — e.g. a custom operating metric — use the metric-name search tool to resolve it to an ID before querying the company-metrics tool.

Currency

Use the company's reporting currency as returned in step 2. Do not default to USD. If the user explicitly asks for a different currency, pass that through to each tool call.

If something is missing

• Company not found → ask the user to confirm spelling; try the legal name.
• A metric returns no value → display —. Never fabricate.
• A tool returns a schema or help error → comply with the introspection call it asks for and retry.
• Chronograph MCP not connected → prompt the user to connect it, or fall back to Model mode.

Model Mode

Read the uploaded Excel file using pandas (data_only=True). Extract from:

Currency and scale: Check the Figures In field on the Overview tab. If 1000 (thousands), divide all financial values by 1,000 before displaying in millions. Note the Local Currency field and include it in the report header.

If commentary is present in the model, use it verbatim (lightly edited for grammar only).

Step 3 — Layout Structure

Both report types share the same structural template. The Financial Performance table must always be directly above the Valuation Summary table in the same column.

┌─────────────────────────────────────────────────────────────────┐
│  HEADER — Logo · Company · Fund · Sector · HQ · Report date    │
│           Tag pills: sector, geography, fund, status           │
├─────────────────────────────────────────────────────────────────┤
│  KPI STRIP (5 cards, full width)                               │
│  LTM Revenue | LTM EBITDA | Enterprise Value | Net Debt | MOIC │
├───────────────────────────┬─────────────────────────────────────┤
│  LEFT COLUMN              │  RIGHT COLUMN                       │
│                           │                                     │
│  Financial Performance    │  Revenue & EBITDA Bar Chart         │
│  Table                    │  (last 4 LTM periods, SVG inline)  │
│                           │                                     │
│  Valuation Summary        │  Investment Returns Table           │
│  Table  ← must stay here  │                                     │
│                           │                                     │
├───────────────────────────┴─────────────────────────────────────┤
│  COMMENTARY (full width, up to 3 columns)                      │
│  Business Update | Valuation Rationale | Discount Rationale    │
├─────────────────────────────────────────────────────────────────┤
│  FOOTER — © {firm_name} | {website} | {confidentiality_label} │
└─────────────────────────────────────────────────────────────────┘

Step 4 — Panel Specifications

Header Strip

• Background: bg_header → bg_primary gradient, 135°
• Logo: if logo_url is set, render <img src="{logo_url}"> at logo_position; maintain clearspace equal to the cap-height of the firm name on all sides
• Logo fallback: if logo_url is blank, render firm name in font_heading, font_heading_weight, 24px, text_primary
• Company name: font_heading, 28px, accent_primary
• Subtitle (fund · sector · HQ · date): font_body, 12px, text_muted
• Tag pills: bg_secondary fill, accent_secondary border, font_body 10px

KPI Strip (5 cards)

• Card: bg_secondary background, 6px border-radius, box-shadow: 0 2px 8px rgba(0,0,0,0.4)
• Value: font_heading, 28px, accent_primary
• Label: font_body, 9px, uppercase, letter-spacing 0.05em, text_primary
• Delta: ▲ +X% in accent_positive · ▼ -X% in accent_negative

Financial Performance Table (LEFT column, top)

Rows (LTM, 3 prior periods + current quarter, vs. PY column):

• Header row: table_header_bg, font_body 9px uppercase, text_primary
• Alternating rows: bg_secondary / bg_primary
• Highlight rows: rgba({accent_primary}, 0.06) background
• Numeric columns right-aligned; label column left-aligned

Valuation Summary Table (LEFT column, directly below Financial Performance)

Prior Quarter vs Current Quarter, plus a delta column (accent_positive / accent_negative).

Revenue & EBITDA Bar Chart (RIGHT column, top)

• Inline SVG only — no external JS or D3
• Side-by-side bars: Revenue in accent_secondary, Adj. EBITDA in accent_negative
• Current quarter Revenue bar: accent_primary
• 4 LTM periods on x-axis; $Xm labels above each bar
• Current quarter label: font_heading, accent_primary; prior: font_body, text_muted
• Y-axis gridlines: dashed, rgba({accent_primary}, 0.08)
• Legend: colour swatches + labels, font_body 9px
• Auto-scale: y-axis max = largest revenue × 1.2; bar heights proportional

Investment Returns Table (RIGHT column, bottom)

• Sort: largest cost first
• Total / Blended row at bottom in font_heading
• Add IRR column if available; omit column if all values null
• Unavailable values: —
• One-sentence QoQ note below table: font_body 9.5px, text_muted

Commentary Section (full width, up to 3 columns)

1. Business Update — verbatim from model if present; AI-generated in MCP mode
2. Rationale for Valuation Conclusion — verbatim from model if present
3. Rationale for Discount to Comps — verbatim from model if present

• Eyebrow label: accent_primary, 9px, uppercase, letter-spacing 0.08em, 1px bottom border in accent_primary
• Body: font_body, 11px, text_muted, line-height 1.6
• Collapse empty columns — never show a blank block

Step 5 — LP Quarterly Update Additions

When report type is LP Quarterly Update:

1. Header eyebrow → LP QUARTERLY UPDATE · Q[N] YYYY
2. Commentary must come from model verbatim — never AI-generate for LP reports
3. Add an optional valuation bridge note below the Valuation Summary table: "EV increase driven by multiple expansion from X.Xx to Y.Yx; EBITDA contribution +$Zm"

Step 6 — CSS Variables & Font Loading

Populate these from the resolved brand tokens (Step 0). Set once in <style>; all panels inherit automatically.

<!-- In <head> -->
<link href="{google_fonts_url}" rel="stylesheet">

<style>
:root {
  --bg-primary:        {bg_primary};
  --bg-secondary:      {bg_secondary};
  --bg-header:         {bg_header};
  --accent-primary:    {accent_primary};
  --accent-secondary:  {accent_secondary};
  --accent-negative:   {accent_negative};
  --accent-positive:   {accent_positive};
  --text-primary:      {text_primary};
  --text-muted:        {text_muted};
  --table-header-bg:   {table_header_bg};
}

body {
  font-family: '{font_body}', sans-serif;
  font-weight: {font_body_weight};
  background: var(--bg-primary);
  color: var(--text-primary);
  margin: 0;
}

h1, h2, h3, h4, .eyebrow, .kpi-value, .moic-value {
  font-family: '{font_heading}', sans-serif;
  font-weight: {font_heading_weight};
}
</style>

Step 7 — Formatting Rules

Step 8 — Output

• File name: [company_name_lowercase_underscored]_[report_type]_[quarter].html
  • Examples: ashworth_health_gp_report.html · ashworth_health_lp_update_q4_2024.html
• Single self-contained HTML file — all CSS in <style>, all data inline
• No external JS — bar chart is inline SVG
• Footer: © {firm_name} | {website} | {confidentiality_label} (omit blank fields)
• Save to the outputs directory and present the file to the user

Error Handling

Data & Brand Checklist (verify before rendering)

Brand

• Brand tokens resolved — from uploaded template or Chronograph defaults
• One-line brand confirmation output to user (or low-confidence question asked)
• CSS variables set; Google Fonts <link> in <head>
• Logo rendered or firm-name fallback applied
• Theme (dark/light) confirmed; contrast ratios verified

Data

• Data source mode determined (Model or MCP)
• Report type determined (GP One-Pager or LP Quarterly Update)
• Currency and scale confirmed (Figures In handling applied if needed)
• Financial Performance table: 4 periods + vs. PY column
• Valuation Summary: PQ vs CQ + delta — in same column as Financial Performance
• Bar chart: 4 periods, bars labelled, y-axis auto-scaled
• Investment Returns: per-investment rows + total row
• Commentary: model text verbatim if available; empty blocks collapsed
• Footer: firm name, website, confidentiality label (omit blanks)
• Single self-contained HTML file, no external JS