fix(telegram): split long messages and sanitize markdown instead of truncating

[konard]

Fixes #137

Root Causes & Fixes

1. Truncated responses (Priority 🔴)

Root cause: handlers.ts simply sliced messages longer than max_message_length (default 4096) with "...", discarding all content after the limit.

Fix: Replaced truncation with splitMessageForTelegram() — a new utility that splits at natural break points (paragraph → line → word boundary) while avoiding cuts inside fenced code blocks. Each part is sent as a separate message.

// Before (handlers.ts)
if (responseText.length > this.config.max_message_length) {
  responseText = responseText.slice(0, this.config.max_message_length - 3) + "..."; // ❌ loses content
}

// After
const parts = splitMessageForTelegram(sanitized, this.config.max_message_length);
for (const part of parts) { await bridge.sendMessage(...); } // ✅ all content delivered

2. Empty rectangles / broken formatting (Priority 🔴)

Root cause: Empty fenced code blocks (```python\n```) render as blank rectangles in Telegram. Unclosed fences cause everything after them to be misrendered.

Fix: New sanitizeMarkdownForTelegram() utility (applied before every send):

• Removes empty code blocks
• Appends missing closing ``` when fence count is odd
• Normalizes 4+-backtick fences (unsupported by Telegram) to ```

New Files

File	Purpose
src/telegram/message-splitter.ts	splitMessageForTelegram(text, maxLength?)
src/telegram/sanitize-markdown.ts	sanitizeMarkdownForTelegram(text)
src/telegram/__tests__/message-splitter.test.ts	10 unit tests
src/telegram/__tests__/sanitize-markdown.test.ts	13 unit tests

Modified Files

File	Change
src/telegram/handlers.ts	Replace truncation with split+sanitize pipeline
src/agent/tools/telegram/messaging/send-message.ts	Apply sanitization before send
src/agent/tools/telegram/messaging/edit-message.ts	Apply sanitization before convert+edit
src/telegram/__tests__/handlers.test.ts	Update test for new splitting behaviour

Test Results

All 23 new tests pass. All 50 existing handlers.test.ts tests pass. Pre-existing failures (@teleton-agent/sdk package unbuilt, schema version mismatch) are unrelated to this PR.

Backward Compatibility

• Short messages (≤ 4096 chars) take the same code path as before — single send, no change in behavior.
• sanitizeMarkdownForTelegram is a no-op on well-formed markdown.
• The max_message_length config option continues to control the per-part limit.

---

This PR was created automatically by the AI issue solver

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Public pricing estimate: $2.304122
• Calculated by Anthropic: $2.304122 USD
• Difference: $0.000000 (+0.00%)

📊 Context and tokens usage:

Claude Sonnet 4.6:

• Max context window: 93.4K / 1M input tokens (9%)
• Max output tokens: 20.6K / 64K output tokens (32%)

Total input tokens: 88.8K + 5.0M cached
Total output tokens: 20.6K output
Cost: $2.140765

Claude Haiku 4.5:

• Context window: 630.1K / 200K tokens (315%)
• Max output tokens: 4.2K / 64K output tokens (7%)

Total input tokens: 69.4K + 560.7K cached
Total output tokens: 4.2K output
Cost: $0.163357

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: sonnet
• Main model: Claude Sonnet 4.6 (claude-sonnet-4-6)
• Additional models:
  • Claude Haiku 4.5 (claude-haiku-4-5-20251001)

📎 Log file uploaded as Gist (1708KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• CI workflows exist but were not triggered for this commit
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag