version: "1.0.0" name: tres-report-create description: Create (generate) any TRES Finance report end-to-end via the tres-mcp MCP server and GraphQL — trigger the export, verify a report row was actually created (guarding against silent failures), poll until done, and return the download link. Trigger this skill on ANY report creation request, including phrases like "create a report", "generate a report", "export", "run a report", "I need a Transaction Ledger / Balances / Reconciliation / Cost Basis / Roll Forward / Staking / Audit report", "pull the data for", "download a report for", "give me a CSV/XLSX of", or whenever the user wants TRES to produce a report file. Use together with tres-report-advisor when the user is unsure which report they need.

TRES Report Create

You create TRES Finance reports programmatically through the tres-mcp MCP server (GraphQL against the BFF). The flow is always: pick the report → trigger the export → poll until DONE → return the presigned link. Optionally download and analyze the file afterwards.

If you are unsure which report the user needs, use the tres-report-advisor skill first, then come back here to generate it.

Workflow

- [ ] Step 0: Clarify report type + date range + currency
- [ ] Step 1: Trigger the export (GraphQL via MCP `execute`)
- [ ] Step 1b: If the report isn't in the tables below, derive it from the schema
- [ ] Step 1c: VERIFY a report row was actually created (poll by exact name) — catches silent failures
- [ ] Step 2: Poll the report query until status == DONE
- [ ] Step 3 (optional): Download + analyze the file

Step 0 — Clarify scope (mandatory before triggering)

Confirm before doing anything:

• Report type (required). Match the user's wording to the tables below. If unsure, discover what's available:

``graphql query { availableReportTypes { name exportType entitiesType llmDescription } } ``

• Date range (if applicable). LEDGER / AUDIT_LOG reports take timestamp_Gte / timestamp_Lte in ISO 8601:
• "Q1 2025" → Gte 2025-01-01T00:00:00Z, Lte 2025-03-31T23:59:59Z
• "2025" → Gte 2025-01-01T00:00:00Z, Lte 2025-12-31T23:59:59Z
• If a date range is required and not given, ask.
• Currency. Default "usd" unless the user says otherwise.
• Analysis goal (only if they want the data analyzed, not just the file). What should the data answer? Default to a general summary (totals, top items, trends). Drives Step 3.

Confirm briefly, e.g. "I'll generate a Transaction Ledger for Q1 2025 as CSV. Starting now."

Step 1 — Trigger the export

Call the GraphQL query that matches the report's entitiesType, passing the export parameters. Run it with the MCP execute tool (validate first with validate_query if unsure).

`entitiesType` → query to call

Report catalog (`exportFormat` values — verified against prod)

† = special handling (see "Reports needing extra parameters" below). The org catalog evolves — availableReportTypes is authoritative. If a requested report isn't in this table, go to Step 1b and derive it from the schema rather than guessing.

Export parameters (always include):

• exportFormat — value from the table above (UPPER_CASE)
• exportName — a descriptive, unique name (e.g. "Transaction Ledger Q1 2025") — you'll match on this in Step 2
• currency — "usd" unless told otherwise
• outputFormat — "CSV" for downstream analysis, or "XLSX" if the user wants a spreadsheet

Add timestamp_Gte / timestamp_Lte for date-ranged reports (LEDGER, AUDIT_LOG).

### ⚠️ GraphQL variable types — get these EXACTLY right (first-run correctness)

The BFF strictly validates variable types. A wrong type returns HTTP 400 and no report is created — and the execute tool surfaces this under an error / error_type field, not the GraphQL errors array, so it is easy to miss. Declare variables with these exact types:

| Variable | GraphQL type | Notes |

|---|---|---|

| exportFormat | String | the UPPER_CASE value |

| exportName | String | unique name |

| currency | String | e.g. "usd" |

| outputFormat | `ReportOutputFormat` | NOT String. Value "CSV" / "XLSX" |

| timestamp_Gte / timestamp_Lte | `DateTime` | NOT String. ISO 8601 value |

| identifier_In, children_Asset_AssetClass_In, children_BelongsTo_In | `[String]` | NOT [ID] |

The params must be passed as variables named exactly exportName, exportFormat, etc. — the BFF reads them from info.variable_values by name. Inline literals or renamed variables ($ef) silently create no report.

Example — trigger a Transaction Ledger export (verified working form):

query($limit: Int, $offset: Int, $timestamp_Gte: DateTime, $timestamp_Lte: DateTime,
      $exportFormat: String, $exportName: String, $currency: String, $outputFormat: ReportOutputFormat) {
  transaction(limit: $limit, offset: $offset, timestamp_Gte: $timestamp_Gte,
              timestamp_Lte: $timestamp_Lte, exportFormat: $exportFormat,
              exportName: $exportName, currency: $currency, outputFormat: $outputFormat) {
    results { id }
    totalCount
  }
}

{
  "limit": 1,
  "offset": 0,
  "timestamp_Gte": "2025-01-01T00:00:00Z",
  "timestamp_Lte": "2025-03-31T23:59:59Z",
  "exportFormat": "BASIC_RAW_TRANSACTIONS",
  "exportName": "Transaction Ledger Q1 2025",
  "currency": "usd",
  "outputFormat": "CSV"
}

The export then generates asynchronously in the background. A 200 with totalCount only confirms the query ran — it does not mean a report was created (verify in Step 1c) and it does not mean the file is ready (poll in Step 2).

Reports needing extra parameters (omit these and the report errors or comes back empty)

Point-in-time balances. For a balance report as of a past date, organizationBalance also accepts balanceDate: DateTime (reconstructs balances at that date — this is the Time Capsule path; passing RAW_BALANCES + balanceDate is promoted to HISTORICAL_BALANCE) and commitId: UUID (pull from a specific commit snapshot). Omit both for current balances.

LOGIN_HISTORY has a different shape

loginHistoryExport takes only exportFormat: String!, exportName: String!, outputFormat: ReportOutputFormat (no currency/limit/timestamps) and returns { success reportId }:

query($exportFormat: String!, $exportName: String!, $outputFormat: ReportOutputFormat) {
  loginHistoryExport(exportFormat: $exportFormat, exportName: $exportName, outputFormat: $outputFormat) {
    success
    reportId
  }
}

Step 1b — Unknown or newly added report? Derive it from the schema (self-service)

availableReportTypes is the source of truth — it lists every report the org supports, including ones not in the tables above. When a requested report has no row here, do not guess: discover the trigger from the live schema using the MCP introspection tools.

- [ ] 1. availableReportTypes -> get this report's exportType + entitiesType + llmDescription
- [ ] 2. Pick the trigger query from entitiesType (Step 1 table). If entitiesType is new/unclear,
         default: transaction (tx / ledger / cost-basis), organizationBalance (balances / assets),
         internalAccount (accounts / metadata).
- [ ] 3. introspect(<query>) -> read the EXACT arg names AND types
- [ ] 4. build_query(<query>) -> canonical variables-as-args template (optional, handy)
- [ ] 5. validate_query(<assembled query>) BEFORE executing
- [ ] 6. Trigger with the 4 export params (+ timestamps only if the query exposes them)
- [ ] 7. Poll; if it errors, apply the error->fix table, then re-trigger
- [ ] 8. On DONE: return the link AND record the working recipe

Principles that make this work first-try:

1. Read the format, never invent it. Use exportType from availableReportTypes verbatim as exportFormat.
2. Derive variable types from `introspect`, not from memory. Declare each variable with the exact type the schema reports. This alone prevents the 400 "used in position expecting type X" failures. Sanity-check the usual ones: outputFormat → ReportOutputFormat, currency → String, timestamp_* → DateTime, list filters → [String].
3. Validate before executing. validate_query catches arg/type/selection mistakes without creating a junk report.
4. Start minimal, then add only what the error demands. Trigger with just the export params first; let the error messages tell you which extra filter a generator requires.

Error → fix table (covers every failure mode seen against prod):

Persist what you learn. Once a new report triggers cleanly to DONE:

• Short term: save the recipe via the MCP memory tool — trigger query, exportFormat, working variable types, and any required filters.
• Durable: propose a new row for this skill's tables (report name, exportFormat, trigger query, required params) and surface it to the user so the skill stays current.

Step 1c — Verify a report row was created (MANDATORY anti-silent-failure check)

Do this immediately after every trigger, before you tell the user anything succeeded. The trigger is a query side effect: if the request is even slightly malformed (inline literals instead of variables, a renamed variable like $ef, a missing required param), the BFF returns 200 with errors: null and silently creates no report. The only reliable way to know it worked is to look the row up by name.

Query for the row using the exact exportName you triggered with:

query($name: String, $ordering: String, $limit: Int) {
  report(name: $name, ordering: $ordering, limit: $limit) {
    results { id name status }
  }
}

{ "name": "<the exact exportName from Step 1>", "ordering": "-created_at", "limit": 1 }

• `results` is empty (no row) → this is a SILENT FAILURE, not "in progress". Do NOT report success. Re-check the trigger against the rules above — params passed as variables (not inline literals), variables named exactly exportName/exportFormat/currency/outputFormat/timestamp_*, correct types — then re-trigger once. If it still creates no row, stop and tell the user it failed silently and could not be created (don't loop forever).
• A row exists → good, the report was created. Continue to Step 2 to poll it to DONE.

Make the exportName unique per run (e.g. append the date or a timestamp) so this lookup matches your row and not an older report with the same name.

Step 2 — Poll until done

Find the report by the exact exportName from Step 1 (most reliable match):

query($name: String, $ordering: String, $limit: Int) {
  report(name: $name, ordering: $ordering, limit: $limit) {
    results { id name status link reportSize exportFormat createdAt }
  }
}

{ "name": "<the exportName from Step 1>", "ordering": "-created_at", "limit": 1 }

Field/arg gotchas (these fail silently — get them right):

• The ordering arg is ordering, not orderBy.
• The download field is link (a presigned S3 URL), not downloadUrl.
• The report-type field is exportFormat, not exportType.
• Do not filter by exportFormat to find your report: the stored value is lowercase (e.g. rollup_breakdown) while you triggered with UPPER_CASE, so an exact filter matches nothing. Filter by name.

Polling loop:

1. Query for the latest report by name.
2. Check status:

• IN_PROGRESS → wait ~60s (e.g. sleep 60 in a shell), then query again.
• DONE → link holds the presigned download URL. Proceed.
• ERROR → the generator rejected the inputs (e.g. a missing required filter). Read the failure message, fix the params, and re-trigger. Do not silently retry the same call.

1. Repeat up to 10 times, waiting ~60s between attempts.

Check BOTH error channels. A report can fail in two ways: (1) the trigger returns HTTP 400 with the problem under error / error_type (and an empty/absent GraphQL errors) — treat any non-empty error as a hard failure, do not report success; (2) the trigger returns 200 but the report later finishes with status: ERROR. Never assume success from a 200 alone.

Small reports finish in seconds; large reports (many assets, wide date ranges, large orgs) can take minutes to hours — set expectations early. Once status == DONE, share the link with the user.

Presenting the download link. Never paste the raw presigned URL into the chat — it's long, ugly, and full of credentials. Always render it as a clean clickable markdown link with friendly anchor text, putting the full URL in the link target:

[Click here to download your report](<the full presigned link>)

You may tailor the anchor text to the report, e.g. [Download your ERP Pre-Sync report (CSV)](<link>). Keep the anchor short and human; the URL itself stays hidden behind it. Mention that the link is presigned and expires after a limited window (~24h).

Step 3 — Download + analyze (optional)

If the user wants the data analyzed (not just the link), download via the presigned URL and inspect it locally with pandas:

import pandas as pd, requests
from io import BytesIO
resp = requests.get("<presigned_url>")
resp.raise_for_status()
df = pd.read_csv(BytesIO(resp.content))
print(df.shape, list(df.columns))
print(df.head())
print(df.dtypes)

Always start with df.head() and df.dtypes, then run the requested analysis (groupby, value_counts, describe, pivot_table). Format money as f"${value:,.2f}" and summarize large frames rather than dumping every row. For deeper per-report analysis guidance, use the tres-report-analyzer skill.

Rules

• ALWAYS confirm report type (and date range, if required) before triggering.
• Match the query to the report's entitiesType using the table above — except the special-case reports above (COST_BASIS_INVENTORY triggers via transaction).
• For any report NOT in the tables, follow Step 1b: introspect the query for exact arg types, validate_query, trigger minimal, fix from the error→fix table, then persist the recipe. Never hand-guess types or filters for an unknown report.
• Variable types are strict: outputFormat is ReportOutputFormat, timestamp_* are DateTime, list filters are [String]. A wrong type = HTTP 400 = no report. Pass params as variables named exactly exportName / exportFormat / etc.
• ALWAYS run Step 1c after triggering. Verify the report row exists by exact exportName before claiming anything worked. An empty lookup means a silent failure (200 + errors: null + no row), not "in progress" — never report success without a confirmed row. Use a unique exportName per run so the lookup matches your row.
• For ROLLUP_BREAKDOWN, you MUST pass identifier_In; for COST_BASIS_INVENTORY, exactly one children_Asset_AssetClass_In — otherwise the report is empty or ERRORs.
• After triggering, verify success on both channels: no error/error_type on the trigger, and status reaches DONE (not ERROR).
• Poll the report query by name, not exportFormat; use ordering (not orderBy) and link (not downloadUrl).
• Present the download link as a clickable markdown link (e.g. [Click here to download your report](<link>)), never the raw presigned URL. Note that the link is presigned and expires after a limited window.
• Default currency to "usd" and outputFormat to "CSV" unless the user specifies otherwise.
• Never truncate or shorten transaction hashes, addresses, report IDs, or identifier_In values — always show/pass them in full.
• This is a read/export skill — NEVER run mutations that modify platform data.