[PATTERN carta-writing-style v0.0.2] [PATTERN etiquette v0.0.6] [PATTERN text v0.0.8] [PATTERN tables v0.0.12] [PATTERN carta-watermark v0.0.10] [PATTERN base v0.1.0]

Consolidating balance sheet

Generates a side-by-side Balance Sheet across every entity under a firm for a single month, matching the "Balance Sheet - consolidating" tab format. The data is pulled live from Carta's DWH via the Carta MCP connector — nothing is embedded in the skill.

This skill runs primarily inside the Claude for Excel add-in. The audience is an accountant working in their workbook, not an engineer running CLI commands.

When to use

Trigger this skill when the user asks for any of the following:

• "Give me the balance sheet of all the entities of <FIRM> for <MONTH>"
• "Consolidating balance sheet for <FIRM>"
• "BS by entity for <FIRM> <MONTH>"
• Any request to replicate the "Balance Sheet - consolidating" tab structure for another firm/period

Do NOT use this skill for:

• Single-entity balance sheets (use a plain trial-balance / GL query instead)
• P&L / income statement requests (use the carta-consolidating-pnl skill)
• Historical multi-period BS (this skill is point-in-time, single month)

UX Rules

This skill ships as a standalone Claude for Excel skill, so the Carta CLI session-start hook does not apply. The following rules — normally enforced globally by the hook — are the skill's own responsibility here:

• Speak plain English to the user. The audience is an accountant. Never surface internal tokens in user-facing prose: MCP server identifiers (claude_ai_Carta_Sandbox), DWH column names (ACCOUNT_TYPE, EFFECTIVE_DATE), UUIDs, raw JSON, SQL, or gate labels. Those belong in Claude's own reasoning, not on screen.
• Currency formatting: positive $X,XXX, negatives ($X,XXX), totals bolded **$X,XXX**, never raw decimals.
• Difference values are absolute — $0 for match, $2,000 for a gap; never -$2,000 in a difference column.
• Status vocabulary when reporting reconciliation outcomes: ✅ Match | ⚠ Mismatch ($X diff) | ❌ Missing in Carta | ❌ Missing in Client Doc.
• Citation links — when pointing to the resulting Excel range, format as [A1:I65](<citation:Balance Sheet - Acme Mar-26!A1:I65>) so the user can jump into the workbook. The range should match the actual written area (column B labels + N entity columns + Total column, by total row count) — adjust per build.
• No environment URLs. This skill builds Excel output, not Carta web links. There is no BASE_URL to construct.

Environment detection (Claude for Excel)

This skill does not call carta auth-status — that command isn't available inside the Excel add-in. Instead, the active Carta environment is detected at Gate 0 from the connected MCP server's prefix (mcp__claude_ai_Carta_<Env>__fetch). Treat the prefix discovery in Gate 0 as the equivalent of "Gate 0: Environment" in Carta CLI skills.

Inputs to collect

Before running, confirm with the user:

1. Firm name — must match a firm reachable from the active Carta MCP context (resolved fuzzily). Example: Acme Ventures.
2. Month — format as YYYY-MM (e.g. 2026-03 for March 2026). The period is cumulative as-of month-end by default, not month-only activity (see references/schema.md for why).

If the user gave both in the request (e.g. "balance sheet of all entities of Acme Ventures for March 2026"), proceed without re-asking.

---

Execution discipline

Execute all gates silently. Do not narrate tool calls, intermediate results, or status updates. Only speak at explicit decision points: Gate 2 (entity scope picker), Gate 5 (pre-build review and approval), Gate 6 (output destination), and Gate 8 (post-action menu).

---

Entry mode — fresh session vs. chained skill

Mandatory telemetry beacon — run this before any skip-gate check: If <SERVER> and <FIRM_UUID> are already in context, call mcp__<SERVER>__set_context(firm_id=<FIRM_UUID>, _instrumentation={"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}) as your very first action. Do not skip this even when all data is already available — it records this skill invocation. If either is unknown, it fires in Gate 0 below.

Before Gate 0, check whether these context variables are already set from an earlier skill call in the same session (e.g. chained from carta-consolidating-pnl):

• <SERVER> — connected Carta MCP server prefix
• <FIRM_NAME> and <FIRM_UUID> — the resolved firm

If both are in context: skip Gate 0 entirely. In Gate 1, skip the contexts:list lookup — but still call mcp__<SERVER>__set_context(firm_id=<FIRM_UUID>, _instrumentation={"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}) to re-anchor the MCP session scope and record this skill invocation, then proceed to fa:list:entities to enumerate entities for Gate 2.

If either is missing (fresh session or cold invocation): run Gate 0 and the full Gate 1 in order.

Do not ask "which firm?" when it is already established from the skill the user just ran.

---

Gate 0: Identify the Carta MCP environment

1. Call refresh_mcp_connectors. Filter servers[] to name matching Carta / Carta (…) / carta with status: "connected". Drop failed.
2. For each connected candidate, probe all three prefix forms in parallel: mcp__claude_ai_Carta__welcome(_instrumentation={"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}) , mcp__carta_production__welcome(_instrumentation={"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}), and mcp__carta__welcome(_instrumentation={"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}). First success = <SERVER>.
3. Don't call any other mcp__<SERVER>__* tool before welcome — every other command is gated and will return a reminder.

If none connected, list failed connectors and stop. If multiple, default to Carta (production). Don't probe every prefix in allowed-tools — only connected ones.

---

Gate 1: Resolve the firm and its entities

1. mcp__<SERVER>__list_contexts(firm_name="<FIRM>", _instrumentation={"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}) to find. Do not use call_tool for list_contexts — call the granular tool directly with _instrumentation as shown. the firm. If multiple matches, present them via AskUserQuestion and confirm.
2. mcp__<SERVER>__set_context(firm_id=<FIRM_UUID>, _instrumentation={"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}) to scope the session. Do not use call_tool for set_context — call the granular tool directly with _instrumentation as shown.
3. call_tool({"name": "fa__list__entities", "arguments": {}, "_instrumentation": {"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}}) to enumerate every entity under the firm.

Prefer the granular tool when the server exposes it — one fewer hop, sidesteps fetch's param-shape quirks:

Granular tool	Generic equivalent
mcp__<SERVER>__list_contexts(firm_name="<entity>")	call_tool({"name": "contexts__list", "arguments": {"firm_name": "<entity>"}})
mcp__<SERVER>__set_context(firm_id="<uuid>")	call_tool({"name": "set_context", "arguments": {"firm_id": "<uuid>"}})

For DWH queries (dwh:execute:query, dwh:list:tables, dwh:get:table_schema) there is no granular equivalent — always go through call_tool({"name": "…", "arguments": {…}}).

Each entity returned from step 3 (fa:list:entities) has the display name, entity_id, and a type field (e.g. entity_type — usually one of FUND, GP, MANAGEMENT_COMPANY, SPV, HOLDING, depending on the firm's structure). Cache the full list — Gate 2 reads from it.

DWH param-name traps — these cost a retry every time:

• dwh:execute:query takes sql:, not query: (the trailing :query in the command name is not the param key).
• dwh:get:table_schema takes table_name:, not table:.

Don't query FUND_ADMIN for entity metadata; the JE table is denormalized and FUND_NAME is on every row.

Done when: the firm is locked and the full entity list (with type, if available) is cached.

---

Gate 2: Choose entity scope

Before pulling data, ask the user which entities to include. The default ("All entities") matches the legacy consolidating-BS behavior; the others let the accountant trim noise from the output before any Excel write happens.

Show a short summary first so the user knows what they're choosing from:

6 entities found under Acme Ventures:

• Funds (3): Acme Fund I, Acme Fund II, Acme Opportunity Fund
• Other (3): Acme Co-Invest (SPV), Acme GP LP (GP), Acme SPV-1 (SPV)

(Group entities by entity_type if the field is available; otherwise group by name pattern — common fund-name tokens are Fund, LP, LLC with Fund in the name. Be conservative — when in doubt, surface the entity in "Other".)

Then ask via AskUserQuestion:

Which entities should the Balance Sheet include?
1 - All entities (6)  ← recommended
2 - Funds only (3) — only fund-type entities
3 - Pick from a list — choose specific entities
4 - Type the names — comma-separated

Handle each branch:

• 1 — All entities → keep every entity from Gate 1.
• 2 — Funds only → filter to entities whose entity_type == "FUND". If entity_type is not present in the API response, fall back to a conservative name pattern (case-insensitive Fund token in the display name). Show the resulting filtered list and confirm with AskUserQuestion before continuing — "Use these 3? Or pick individually?"
• 3 — Pick from a list → present a multi-select AskUserQuestion where each option is one entity name. Require at least one selection.
• 4 — Type the names → accept a comma-separated list as free text. Fuzzy-match each token against the cached entity names; show the resolved list and any unmatched tokens; confirm via AskUserQuestion before continuing. If any tokens are unmatched, loop until they're resolved or the user removes them.

Lock the chosen list as <entity_scope> and use it in Gate 3.

Wait for the user to confirm before continuing.

If the user picks an option that resolves to zero entities, stop and ask them to re-pick — never run the BS query with an empty entity scope.

Done when: <entity_scope> is a non-empty, user-confirmed list of entity display names from the Gate 1 cache.

---

Gate 3: Pull journal entries

The schema and sign conventions for the Carta DWH journal-entries table are documented in references/schema.md. Load that file now and apply its rules.

Default query — cumulative as-of (this is what balances):

SELECT FUND_NAME, FUND_UUID, ACCOUNT_TYPE, ACCOUNT_NAME, SUM(AMOUNT) AS BALANCE
FROM <journal_entries_table>
WHERE FIRM_ID = '<firm_uuid>'
  AND EFFECTIVE_DATE <= '<YYYY-MM-DD month_end>'
  AND ACCOUNT_TYPE BETWEEN '1000' AND '3999'
  AND FUND_NAME IN (<comma-separated entity_scope names, quoted>)
GROUP BY 1, 2, 3, 4
ORDER BY FUND_NAME, ACCOUNT_TYPE

Use a FUND_NAME IN (...) clause only when <entity_scope> is not the full Gate 1 list (i.e. the user picked options 2, 3, or 4). When the user picked "All entities" (option 1), omit the FUND_NAME filter so the query plan stays simple.

Single-quote each name in the IN (...) clause and escape any embedded single quotes by doubling them (e.g. O'Connor → 'O''Connor').

DWH result formatting — keep the response small

For monthly / per-account / per-period activity queries that return more than ~50 rows, request format: "ndjson" and bucket the result into a blob (store_blob / blobs.setJSON / the runtime equivalent) before working with it. Do not paste large query results back into the conversation — they trigger context_snip, which can compress your Gate 4 / Gate 5 checks out of user visibility.

Supported format values for dwh:execute:query:

• markdown — best for small (≤50 row) results that you want to show the user.
• ndjson — best for large results processed by code/agent.
• csv is NOT supported. Do not try it.

Run via call_tool({"name": "dwh__execute__query", "arguments": {"sql": "..."}, "_instrumentation": {"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}}). SELECT-only.

Period-only variant (EFFECTIVE_DATE BETWEEN <month_start> AND <month_end>) — run only if the user explicitly asks for "this month's activity" or "period-only." Warn them first: it will not balance, because BS accruals booked during the month have offsetting P&L entries excluded by the ACCOUNT_TYPE filter.

If the result set is large, store it in a session blob (e.g. blobs.setJSON("bs_data", ...)) instead of carrying it in the prompt.

Resolve the presentation currency (<fund_currency>)

The number format in references/formatting.md is built from <fund_currency> — resolve it here, do not assume USD:

1. Probe the journal-entries table for a currency column: call_tool({"name": "dwh__get__table_schema", "arguments": {"table_name": "<journal_entries_table>"}, "_instrumentation": {"plugin": "carta-investors", "skills": ["carta-consolidating-balance-sheet"]}}). If it exposes a currency column (e.g. CURRENCY, REPORTING_CURRENCY, FUND_CURRENCY), add SELECT DISTINCT <currency_col> scoped to <entity_scope> and read the value(s).
2. If exactly one distinct currency comes back, set <fund_currency> to it.
3. If the column doesn't exist, or the scope spans multiple currencies (a consolidating BS that mixes currencies cannot be summed into one Total column — see the no-cross-currency rule), ask the user via AskUserQuestion: "What presentation currency should I use for this consolidating balance sheet?" Store the answer as <fund_currency>.

Never silently default to USD.

Done when: the period dataset is loaded (covering only Asset / Liability / Equity accounts) — scoped to <entity_scope> — and <fund_currency> is resolved.

---

Gate 4: Classify accounts

Classify each row by the leading digit of ACCOUNT_TYPE, per references/schema.md:

• 1xxx → Assets — keep BALANCE as-is
• 2xxx → Liabilities — multiply BALANCE by -1 for positive display
• 3xxx → Equity — multiply BALANCE by -1 for positive display

Sort within each section by ACCOUNT_TYPE ascending. Use ACCOUNT_NAME directly as the row label — do not rename or consolidate accounts.

Done when: every account has a section and a positive display value.

---

Gate 5: Pre-build review

Before touching the user's workbook, show a plain-English preview so the accountant can sanity-check the build and edit if anything looks off. This is the pre-flight checkpoint — no Excel changes happen until the user explicitly confirms.

Present a short, scannable summary. Echo back the scope choice from Gate 2 so the user can see exactly what's in and what's out:

Ready to build the Balance Sheet — please review.

• Firm: Acme Ventures
• Period: as of March 31, 2026 (cumulative)
• Entity scope: Funds only (3 of 6) — Acme Fund I, Acme Fund II, Acme Opportunity Fund (Excluded: Acme Co-Invest, Acme GP LP, Acme SPV-1.)
• Accounts found: 22 Assets, 11 Liabilities, 7 Equity (40 total)
• Sheet name: Balance Sheet - Acme Mar-26

If the user picked "All entities" in Gate 2, render the scope line as:

• Entity scope: All entities (6) — Acme Fund I, Acme Fund II, Acme Co-Invest, Acme GP LP, Acme Opportunity Fund, Acme SPV-1

Then ask with AskUserQuestion:

1 - Build it  ← recommended
2 - Change the firm or period
3 - Change the entity scope
4 - Cancel

Handle each branch:

• 1 — Build it → proceed to Gate 6.
• 2 — Change firm or period → return to Inputs, re-run Gates 1–4 with the new values, then present this review again.
• 3 — Change the entity scope → return to Gate 2 (the entity-scope picker). On confirm, re-run Gates 3–4 with the new scope, then present this review again.
• 4 — Cancel → stop the skill cleanly.

Loop until the user picks "Build it" or cancels. Never write to Excel based on inferred intent.

Do not surface internal field names (ACCOUNT_TYPE, FUND_NAME, EFFECTIVE_DATE) or UUIDs in this review — translate to plain accountant language ("accounts", "entities", "as of …").

Hard rule: no workbook-write tool (Excel-add-in cell write, execute_office_js that mutates state, write_workbook.py, or any equivalent) runs before this gate's AskUserQuestion returns the user's explicit "Approve and write" choice. If you catch yourself about to call a workbook-write tool without that approval recorded, stop and run this gate first.

Done when: the user has confirmed the build, with their chosen firm, period, and entity scope locked in.

---

Gate 6: Decide the output destination

This skill is designed to run inside the Claude for Excel add-in. Before writing anything, decide whether to write into the user's currently open workbook or to create a new one.

1. Check for an active workbook. Use the Excel add-in's "active workbook" / "current workbook" tool (whatever name the add-in exposes at runtime) to see if there is a workbook open in front of the user.
2. If a workbook is open:
   • Read the workbook name and the full list of existing sheet/tab names.
   • Scan for a matching existing tab (COA label detection). For each existing tab, read all non-empty cell values from column B via execute_office_js. Compare them against the ACCOUNT_NAME values in the Gate 4 dataset. A tab is a COA-label match if ≥ 5 account labels from the current query appear in that tab's column B. An exact-name match is a tab whose name equals the proposed sheet name (Balance Sheet - <FIRM-SHORT> <MMM-YY>). Check for both — a renamed tab can match on labels even when its name differs.
   • If a matching tab is found (exact-name or COA-label): ask the user via AskUserQuestion, naming the matched tab explicitly:

     "I found an existing tab <matched_tab_name> that appears to contain balance sheet data for this firm. What would you like to do?" Options:

     • "Update the existing <matched_tab_name> tab" — clear and rebuild it in place. ← recommended
     • "Create a new tab instead" — adds a new tab named Balance Sheet - <FIRM-SHORT> <MMM-YY> (with a numeric suffix like (2) if that name also already exists; truncate to Excel's 31-character limit after suffixing).
     • "Cancel" — stop the skill.

   • If no matching tab is found: ask the user whether you may add a new tab to the workbook. Example:

     "You have <workbook name>.xlsx open. May I add a new tab called Balance Sheet - Acme Mar-26 to it?" Options:

     • "Yes, add the tab here" — proceed with the active workbook.
     • "No, create a new workbook instead" — create a fresh workbook.
     • "Cancel" — stop the skill.

   • If the user picks "Update existing tab": clear the matched tab's used data range before writing:

     const sheet = context.workbook.worksheets.getItem("<matched_tab_name>");
     sheet.getUsedRange().clear();
     await context.sync();
     

     Then proceed to Gate 7 using <matched_tab_name> as the target sheet (the sheet already exists — do not call create_sheet).
3. If no workbook is open:
   • Create a new workbook silently. Name it Balance Sheet - <FIRM-SHORT> <MMM-YY>.xlsx (e.g. Balance Sheet - Acme Mar-26.xlsx). The first sheet is the consolidating Balance Sheet tab.
   • Tell the user, in one sentence, that you created a new workbook because nothing was open.
4. If the user cancels or denies edit permission for the active workbook and picks "Cancel": stop the skill cleanly. Don't fall back silently to creating a new file.

Lock the chosen <destination workbook> and <sheet name> and use them through Gate 7 and Gate 8.

Done when: the destination workbook + target sheet name are known and the user has explicitly consented to any edit to a pre-existing workbook.

---

Gate 7: Build and brand the output sheet

Approval-recorded check (run FIRST, before any write tool)

Before calling execute_office_js with state-mutating code, setValues, write_workbook.py, or any other workbook-write tool, look back at your tool history. Find the most recent AskUserQuestion you sent. Does its answer literally include "Build it" (the Gate 5 build-approval)? If NO, Gate 5 did not pass — send the Gate 5 approval menu now and wait for the explicit answer.

Do not interpret upstream answers as approval. An entity-scope response from Gate 2, a destination response from Gate 6, or any prior AskUserQuestion whose answer is not literally a build-approval choice does NOT clear this gate.

Gate 7 requires AT LEAST three separate execute_office_js calls (excel-addin runtime)

The most common failure mode is bundling cell writes + formatting + logo into one writeSheet(...) function — the model writes the cells, returns, and forgets the logo. Do not combine the cell-write call with the brand block in a single office.js block.

• Call 1: cell values, formulas, formatting. One execute_office_js. Return.
• Call 2: logo on the BS tab via the verbatim brand block below.
• Call 3 (verification, LAST): load shape names on the BS tab, confirm CartaLogo exists.

Returning from Call 1 does NOT finish Gate 7. The verification call must appear in your tool history before Gate 8.

Read references/formatting.md now and apply its layout exactly. That file covers sheet name, header rows, section headers, subtotal/grand-total rows, the Total column, number formats, column widths, font, and the "do not include" list.

Also read references/branding-and-header.md for the standard 4-row metadata band (B1 firm / B2 descriptive title like "Consolidating Balance Sheet · As of Mar-26" / B3 source / B4 other context) and Carta logo placement (column E, rows 1–3 height). If formatting.md and branding-and-header.md disagree on where section headers start (e.g. formatting.md says B5 = Assets), the branding spec wins — shift sections to start at row 6 or below so rows 1–4 stay reserved for metadata.

Brand block — verbatim, run AFTER the cell writes (DO NOT SKIP)

The output sheet is not "built" until it carries a CartaLogo shape. Use the verbatim brand block from references/branding-and-header.md — substitute <TAB_NAME> with the BS tab name. Asset access via blobs.getText("assets/powered_by_carta.b64.txt") — NOT Read or filesystem paths.

Brand-verification call (REQUIRED, observable). Run this as a separate execute_office_js call before proceeding to Gate 8:

const sheet = context.workbook.worksheets.getItem("<BS_TAB_NAME>");
sheet.shapes.load("items/name");
await context.sync();
return sheet.shapes.items.map(s => s.name);

The result must include "CartaLogo". If not, re-run the brand block above and re-verify. Do not proceed to Gate 8 until this verification returns CartaLogo. The verification call is observable evidence; without it in your tool history, Gate 7 branding is not complete.

Two critical reminders from that reference:

• Write the Total-column header (e.g. Mar-26) with numFmt = "@" applied before setting the value. Otherwise Excel parses it as a date serial.
• Every total is a =SUM(...) formula, never a hardcoded number. The Total column sums across entities; section subtotals sum down the section.

Empty entities: if an entity in <entity_scope> has no JE activity through the as-of date, still include the column with zeros — not blanks.

Sheet-write hard rules

• range.merge(true) discards values in trailing cells. Never merge cells whose contents you still need. To add a header row above existing sub-headers, insert a new row first (sheet.getRange("5:5").insert(...)) and write the merged labels into the inserted row — do not merge over a row that already holds sub-headers.
• Two-row header pattern when bucketing by period: row N = one merged cell per period (e.g. Mar-26, Apr-26); row N+1 = sub-headers if any. Never write the period label and any sub-headers into the same row.
• The date-serial trap (already called out above for the Total-column header) applies to every cell where you write a month/period string. Set numFmt = "@" before the value, or apply numberFormat: "mmm yyyy" to a real date in the same write.

Done when: the sheet exists with one column per entity in <entity_scope>, all account rows, all subtotals, and the grand-total row, all driven by formulas.

---

Gate 8: Verify and report

Gate 8 precondition (DO NOT SKIP). Before sending the report text below, scan your tool history. Three anchors MUST be present in that order:

1. An AskUserQuestion whose answer included "Build it" (Gate 5 build-approval).
2. A sheet.shapes.addImage(base64) call for the BS tab — Gate 7 branding.
3. The branding-verification execute_office_js whose result included "CartaLogo" — Gate 7 verification.

If any anchor is missing, you have skipped a gate. Do NOT report build success when no shapes.addImage call appears in your tool history. STOP, go back, run the missing gate, then return here.

After writing:

1. Read back the Total Assets and Total Liabilities and Equity rows. They should match per entity (BS must balance).
2. Flag any entity where |Total Assets - Total Liabilities and Equity| > 1000 (more than $1k off in raw dollars). Imbalance usually means the COA has accounts that didn't classify cleanly into Assets, Liabilities, or Equity.

Report structure:

Lead with a one-line confirmation, then a Key tie-outs block, then the detail. Status vocabulary: ✅ Match, ⚠ Mismatch ($X diff).

The Balance Sheet is ready in Acme Q1 Reporting.xlsx — tab Balance Sheet - Acme Mar-26.

Key tie-outs (per-entity foot, Assets vs Liabilities & Equity):

Entity	Total Assets	Total Liab & Equity	Difference	Status
Acme Fund I	$48,210,300	$48,210,300	$0	✅ Match
Acme Fund II	$22,815,900	$22,815,900	$0	✅ Match
Acme Co-Invest	$6,440,775	$6,440,775	$0	✅ Match
Acme GP LP	$1,228,100	$1,228,100	$0	✅ Match
Acme Opportunity Fund	$13,902,440	$13,902,440	$0	✅ Match
Acme SPV-1	$3,118,650	$3,118,650	$0	✅ Match

6 entities checked. 6 matched, 0 mismatched.

51 accounts written across 6 entity columns + Total column.

If any entity does not foot, surface its row with status ⚠ Mismatch ($X diff) and a short note pointing the user at the likely cause (unclassified account, COA digit drift, period-only variant ran by mistake).

Then offer the post-action menu:

The next-step menu MUST be a single AskUserQuestion call with the options below as options entries. Never render them as a numbered markdown list, a bulleted list, or inline prose — bare-text menus break the chooser UI in Claude for Excel and force the user to type the number. The ← recommended marker goes inside the description field of one option, not as a suffix on the label.

1. Build the P&L for the same firm and period ← recommended
2. Build the Balance Sheet for a different period
3. I'm done

When the user selects an option, immediately invoke the corresponding skill via Skill('<skill-name>') BEFORE doing any work. Do not freelance the output — load the downstream skill's SKILL.md so its gates, layout spec, branding rules, and approval flow apply. Routing:

Option	Skill to invoke
1 — Build the P&L	Skill('carta-investors:carta-consolidating-pnl')
2 — Build the Balance Sheet for a different period	Skill('carta-investors:carta-consolidating-balance-sheet') re-entry with the new period
3 — I'm done	No invocation; close cleanly

Done when: sheet exists, Key tie-outs reported with status per entity, post-action menu rendered.

---

Schema discovery

The skill queries the Carta DWH journal-entries table. If column names are needed, look up the table via the Carta MCP DWH schema command once at Gate 0 — production schema is canonical. Don't embed column listings inline; the DWH contract can drift.

Error handling

Symptom	Likely cause	What to tell the user
No Carta MCP tool present	The Carta connector isn't enabled in this Claude for Excel session	"I can't see your Carta connector. Open Settings → Connectors in Claude, enable Carta, then ask me again."
contexts:list returns no firm	Firm name typo, or the user's MCP context doesn't include this firm	Echo back the name and ask for confirmation or an alternative spelling. Don't silently pick a near-match.
contexts:list returns multiple firms	Common short names match several firms	Show the candidates via AskUserQuestion and let the user pick.
Query returns 0 rows	No JE activity for the firm through the as-of date	"I didn't find any journal entries for this firm through Mar 31, 2026. Double-check the period or confirm books are open for that month."
Query times out	DWH load, or unusually large date range	Tell the user the query is slow and offer to retry with the same parameters — never auto-retry.
BS doesn't balance per entity	COA accounts not in 1xxx/2xxx/3xxx, or period-only variant ran by mistake	Surface the per-entity imbalance in Gate 8's Key tie-outs table. Suggest reviewing the unclassified accounts.
Auth / permission error from MCP	The user's Carta session expired or lacks DWH access	Tell them to reconnect Carta in Settings → Connectors. Do not retry automatically.

Never auto-retry a failed command. Always surface the failure and let the user decide whether to retry.

---

Do NOT

• Don't skip Gate 2 by inferring entity scope from chat. User must explicitly pick — even if the prompt says "all entities", echo the resolved list and confirm.
• Don't run BS query with empty entity scope. Zero entities → stop, ask user to re-pick.
• Don't include Eliminations, Topside Adjustments, or prior-period columns. Only non-entity column is Total.
• Don't rename accounts. Use ACCOUNT_NAME from JE verbatim.
• Don't run period-only variant without explicit opt-in and a "won't balance" warning.
• Don't claim success without re-reading and verifying per-entity balance in Gate 8.
• Don't skip branding. Gate 8 report must not run until the tab carries CartaLogo on column E — see references/branding-and-header.md.
• Do NOT freeze panes on this tab.