openclaw - 💡(How to fix) Fix bug(telegram): media captions can exceed 1024 after HTML rendering

Summary

Telegram media caption splitting currently checks the raw caption length before Telegram HTML rendering. A caption can be <= 1024 raw UTF-16 code units, be accepted as an inline caption by OpenClaw, then expand beyond Telegram's 1024 caption limit after HTML escaping / Markdown rendering.

This can cause Telegram media delivery failures or confusing cut/truncated delivery around media replies. It is related to the Telegram HTML/caption transport area, but is distinct from the existing angle-bracket truncation bug.

Repro

Raw caption:

"&".repeat(300)

Observed planning today:

• raw caption length: 300
• rendered Telegram HTML caption length: 1500, because & becomes &
• current splitTelegramCaption() keeps it inline because it only sees 300 <= 1024

Another repro:

"[" + "link".repeat(20) + "](https://example.com/" + "x".repeat(900) + ")"

Observed:

• raw length: 1024
• rendered Telegram HTML length: 1035
• still selected as caption before rendering

Expected

OpenClaw should decide whether to send media text as a caption after rendering to the actual Telegram HTML payload. If rendered caption length exceeds 1024, media should be sent without caption and the text should be sent as follow-up text.

Local mitigation tested

A local hot patch checks rendered caption length after renderTelegramHtmlText() and moves oversized rendered captions to follow-up text. Runtime import/config/restart passed on v2026.5.12.

Candidate fix shape

Add a rendered-caption splitter/helper, e.g.:

• split raw caption by existing raw >1024 guard
• render candidate caption to Telegram HTML
• if rendered HTML length >1024, return followUpText
• reuse this in both direct Telegram send and reply media delivery paths

Environment

• OpenClaw: 2026.5.12
• Channel: Telegram
• Path observed in package build: extensions/telegram/src/caption.ts, extensions/telegram/src/send.ts, extensions/telegram/src/bot/delivery.replies.ts