A downloadable CSV is not proof of a correct export. A comma in a note can split a column, a spreadsheet can rewrite an identifier, and user-entered text can become a formula. Acceptance should cover who receives which records, how values are represented, and what will read the file.
The historical Russian client brief Tg бот на Django requests a Telegram bot, a Django administration interface, and CSV export. It was published November 4, 2025, and appeared closed and archived when checked November 8, 2026. It establishes a real export request, not a BitBear delivery case. Its linked Figma specification was not inspected. The personal-record scenario, policies, and checks below are additional design proposals, not the client's complete specification.
Define the export boundary before its format
For a personal note bot, start with private-chat exports of the requesting user's authorized, active records. List necessary columns such as ID, creation time, title, and note. Do not attach phone numbers, internal logs, tokens, or another user's data by default.
Show filters, record count, and generation time. Define whether date selection uses creation time or a business date. If records change while an export runs, require a consistent snapshot or disclose the cutoff and limitations. Combining pages from different moments must not masquerade as an exact result.
Check authorization both when requested and before delivery. Send only to an authorized recipient; a command posted in a group does not justify posting its attachment back there. For no results, return an explicit message or headers-only file. Report capacity limits instead of silently truncating rows and claiming completeness. A file is a copy, not ongoing synchronization or a proven restorable backup.
Serialize CSV instead of joining strings
RFC 4180 documents a common format in which fields containing commas, quotes, or newlines require quoting, and embedded double quotes are doubled. It is Informational, not a guarantee that every program interprets files identically.
Test a note containing a comma, a quote, and two lines of text. After parsing, it must remain one field. Data rows must have the expected number of columns. Count parsed records rather than raw newline characters, because a field can span lines.
Choose UTF-8, a delimiter, column order, timestamp format, and a null policy. Use invented Chinese, English, Russian, and emoji samples. If consumers must distinguish a missing value from an intentionally empty string, add an explicit marker or choose an exchange format that preserves the distinction. An empty cell does not automatically retain every business meaning.
Correct syntax does not prevent type conversion
CSV does not declare strong column types. An ID such as 00123, or a long string of digits, can be an identifier rather than a quantity. CSV double quotes are not a contract forcing a spreadsheet column to remain text.
Microsoft's leading-zero and large-number guidance explains conversion risks and how to import relevant columns as Text. Include instructions for the chosen spreadsheet application. Changing display formatting after the original value has been lost cannot reconstruct it.
Do not turn an identifier into a formula to preserve its appearance. If users need a file that opens without configuration and retains column types, separately evaluate a workbook with explicitly typed text cells. That requires real library, runtime, and compatibility support. Renaming .csv to .xlsx does not provide it.
CSV quoting is not formula protection
OWASP CSV Injection explains that spreadsheet applications may interpret untrusted cell contents as formulas. Common defenses can fail after saving and reopening, and no sanitization method suits every consumer.
Consider the harmless test note =1+1. It is business text, not an instruction to calculate when viewing an export. Treat names, titles, notes, and custom fields as untrusted. Checking only the first column, or removing real names, does not address the problem.
Separate two delivery profiles:
- Machine exchange: preserve agreed values and structure for a controlled parser. Explicitly label it as unsuitable for direct spreadsheet viewing.
- Human viewing: choose a text-handling policy for a named application and version. Test opening, saving, closing, and reopening. If protection changes exported text, disclose the change and leave original business records intact.
Do not indiscriminately delete leading characters or promise that adding one prefix everywhere is permanently safe. Choose the consumer, then validate the specific approach. If the viewing path cannot reliably keep text from becoming formulas, defer one-click spreadsheet downloads and retain controlled previews, or separately implement and test a typed file format.
Where BotFatherV2 can help
As checked November 8, 2026, BotFatherV2 supports AI-assisted creation and modification of Telegram bots. Its requested /en-US entry was inaccessible; capability checks used the accessible English homepage and Build with AI documentation.
User-isolated persistent tools are a documented starting direction. Describe personal records and export policies, then verify the current SDK's query, encoding, file-generation, and attachment-delivery support. This does not establish ready-made CSV, Excel, or Django modules.
The fixed SDK excludes arbitrary npm packages and Node/Bun APIs; the initial release does not add paid-AI-dependent runtime capabilities. If file delivery is unsupported, start with authorized record previews and report the gap instead of claiming a successful export. The original client's Django, PostgreSQL, and Celery stack needs independent development. Imports are limited to existing Telegram Serverless bots.
BotFatherV2 is independent of Telegram's official BotFather. Verify storage and delivery separately: Telegram accepting attachments does not prove arbitrary file-generating code can run.
A reusable brief and acceptance checklist
Design a private personal-note bot. Export only the current user's authorized active records with fixed ID, creation-time, title, and note columns. Show filters, cutoff, record count, and completeness status. First verify SDK support for consistent queries, UTF-8, CSV serialization, file generation, and private delivery; otherwise provide record previews only. Separate machine exchange from spreadsheet viewing. Double quotes alone are not formula protection; test the viewing policy in the chosen application. Do not alter source records, export another user's data or secrets, silently truncate results, or add Django administration and automatic synchronization.
At minimum, check that:
- Commas, quotes, newlines, Chinese text, and emoji round-trip according to the selected contract, without shifted columns.
00123and long IDs remain character-for-character identical in the agreed import procedure.- Harmless
=1+1remains text in the spreadsheet-viewing path, including after save and reopen. - Counts and stable IDs reconcile with the same-cutoff query; concurrent edits follow the published snapshot rule.
- User B cannot change parameters to export A's records. Revoked access or failed delivery does not yield a false success message.
- Empty, oversized, and partially failed exports are distinguishable; partial files are never labeled complete.
These are proposed acceptance conditions, not features of a bot generated or tested for this article. For amount parsing and financial totals, see the separate expense-bot guide. This article concerns preserving meaning between business records, files, and reading software.
Sources and verification
- FL.ru, Tg бот на Django, Russian buyer brief, published November 4, 2025. Checked November 8, 2026: closed and archived. Only the public request for CSV export informs the scenario.
- Y. Shafranovich, RFC 4180, October 2005, English Informational document. The entry redirects to its RFC information page; checked November 8, 2026.
- Timo Goosen and Albinowax / OWASP, CSV Injection, maintained English material without a single publication date. The older entry redirects here; checked November 8, 2026.
- Microsoft, Keeping leading zeros and large numbers, official English support documentation without a single publication date; checked November 8, 2026.