This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 6

Comment:
`RTL_MARK` is declared but never used

The constant `RTL_MARK` is defined here but is never referenced anywhere in the file. The PR description says it "prepends Unicode RLM mark to RTL text in `sendMessageDraft`", but no such prepending logic exists — there is no RTL detection, and `RTL_MARK` is never passed to `sendDraftTransportPreview`, `sendMessageTransportPreview`, or any text construction path.

As a result, this fix is a no-op and issue #32852 remains unresolved.

The actual implementation should detect RTL text (e.g. using a regex against RTL Unicode ranges) and prepend `RTL_MARK` to the rendered text before it is sent. For example, inside `sendOrEditStreamMessage` after the text is rendered:

```ts
const isRtl = /[\u0591-\u07FF\uFB1D-\uFDFD\uFE70-\uFEFC]/.test(renderedText);
const finalText = isRtl ? RTL_MARK + renderedText : renderedText;
```

and then use `finalText` in place of `renderedText` for the send calls.

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 146-149

Comment:
**Inconsistent indentation on `await` call**

The `const textWithRTLM` declaration uses 8-space indentation (matching the `if` body), but the `await params.api.editMessageText(...)` call that follows is indented only 6 spaces, breaking the block's consistent indent level and likely flagging a linter/formatter error.

```suggestion
        const textWithRTLM = isRTL(renderedText) ? "\u200F" + renderedText : renderedText;
        await params.api.editMessageText(chatId, streamMessageId, textWithRTLM, {
          parse_mode: renderedParseMode,
        });
```

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 147

Comment:
The `await` call is indented to 6 spaces but should be 8 spaces to align with the `const` declaration on the line above, which is inside the same `if (renderedParseMode)` block.

```suggestion
      await params.api.editMessageText(chatId, streamMessageId, textWithRTLM, {
```

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 232

Comment:
**`trimEnd()` called multiple times on the same expression**

`rendered.text.trimEnd()` is evaluated three times in this line: once for the `isRTL()` check, once for the truthy branch, and once for the falsy branch. Store the trimmed value in a variable first to avoid redundant allocations:

```suggestion
    const trimmedText = rendered.text.trimEnd();
    const renderedText = isRTL(trimmedText) ? "\u200F" + trimmedText : trimmedText;
```

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 119

Comment:
**RTL regex may miss Arabic Presentation Forms**

The range `[\u0590-\u08FF]` covers Hebrew and the primary Arabic Unicode blocks, but misses Arabic characters that appear via Presentation Forms:

- `\uFB50-\uFDFF` — Arabic Presentation Forms-A
- `\uFE70-\uFEFF` — Arabic Presentation Forms-B

These blocks are used in some legacy and display contexts. Text containing only Presentation Form Arabic characters would not be detected as RTL, leaving the fix incomplete for those edge cases.

Consider expanding the range to include them:

```suggestion
  const isRTL = (text: string) => /[\u0590-\u08FF\uFB1D-\uFDFD\uFE70-\uFEFC]/.test(text); // Hebrew, Arabic and Arabic Presentation Forms
```

Additionally, since `isRTL` does not close over any variables from `createTelegramDraftStream`, it could be extracted as a module-level constant to avoid recreating the function on every factory call.

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 232

Comment:
`rendered.text.trimEnd()` is evaluated three times in this line — once in the `isRTL()` check, once in the truthy branch, and once in the falsy branch. This causes redundant string allocations.

Store the trimmed value in a variable first:

```suggestion
const trimmedRenderedText = rendered.text.trimEnd();
const renderedText = isRTL(trimmedRenderedText) ? "\u200F" + trimmedRenderedText : trimmedRenderedText;
```

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 119

Comment:
The RTL regex range `[\u0590-\u08FF]` covers Hebrew and primary Arabic blocks but misses Arabic Presentation Forms, which are used in some legacy and display contexts:

- `\uFB50-\uFDFF` — Arabic Presentation Forms-A
- `\uFE70-\uFEFF` — Arabic Presentation Forms-B

Text containing only these presentation-form characters would not be detected as RTL. Consider expanding the regex to include them:

```suggestion
const isRTL = (text: string) => /[\u0590-\u08FF\uFB1D-\uFDFD\uFE70-\uFEFC]/.test(text);
```

Additionally, since `isRTL` doesn't close over any variables from `createTelegramDraftStream`, it could be extracted as a module-level constant to avoid recreating the function on every factory call.

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 232

Comment:
`rendered.text.trimEnd()` is evaluated three times on this line — once in the `isRTL()` check, once in the truthy branch, and once in the falsy branch. This causes redundant string allocations.

Store the trimmed value once instead:

```suggestion
    const trimmedText = rendered.text.trimEnd();
    const renderedText = isRTL(trimmedText) ? "\u200F" + trimmedText : trimmedText;
```

How can I resolve this? If you propose a fix, please make it concise.

This is a comment left during a code review.
Path: src/telegram/draft-stream.ts
Line: 119

Comment:
The RTL detection regex `[\u0590-\u08FF]` covers Hebrew and the primary Arabic Unicode blocks, but misses Arabic Presentation Forms used in some legacy and display contexts:

- `\uFB50-\uFDFF` — Arabic Presentation Forms-A
- `\uFE70-\uFEFF` — Arabic Presentation Forms-B

Text containing only these presentation-form characters would not be detected as RTL. Consider expanding the regex to include these ranges:

```suggestion
  const isRTL = (text: string) => /[\u0590-\u08FF\uFB50-\uFDFF\uFE70-\uFEFF]/.test(text); // Hebrew, Arabic ranges, Arabic Presentation Forms
```

How can I resolve this? If you propose a fix, please make it concise.