inline code`
+- Fenced code blocks (` ```lang\n...\n``` `) → `...` +- `[text](url)` → `text` + +#### Scenario: Headers converted to bold +- **WHEN** input contains `## Market Summary` +- **THEN** output SHALL contain `Market Summary` + +#### Scenario: Bold and italic preserved +- **WHEN** input contains `**important** and *emphasis*` +- **THEN** output SHALL contain `important and emphasis` + +#### Scenario: Code blocks preserved +- **WHEN** input contains a fenced code block with content `const x = 1` +- **THEN** output SHALL contain `
const x = 1` with HTML entities escaped inside + +#### Scenario: Links converted +- **WHEN** input contains `[Google](https://google.com)` +- **THEN** output SHALL contain `Google` + +### Requirement: HTML entity escaping +The converter SHALL escape `<`, `>`, and `&` in non-code text to `<`, `>`, and `&` before applying formatting transformations. Code blocks and inline code SHALL also have their content escaped to prevent HTML injection. + +#### Scenario: Angle brackets escaped +- **WHEN** input contains `price < 100 && price > 50` +- **THEN** output SHALL contain `price < 100 && price > 50` + +#### Scenario: Code content escaped +- **WHEN** input contains `` `
<div>hello</div>`
+
+### Requirement: TelegramConnector uses HTML parse mode
+`TelegramConnector.send()` SHALL convert message text through `markdownToTelegramHtml()` and send with `parse_mode: 'HTML'`. If the Telegram API rejects the HTML (parse error), the system SHALL fall back to sending as plain text without `parse_mode`.
+
+#### Scenario: Formatted message sent
+- **WHEN** `send({ text: '## Title\n**bold**' })` is called
+- **THEN** `bot.api.sendMessage` SHALL be called with `parse_mode: 'HTML'` and converted HTML content
+
+#### Scenario: HTML parse failure fallback
+- **WHEN** Telegram API rejects the HTML content
+- **THEN** the system SHALL retry `sendMessage` without `parse_mode` (plain text)
+
+### Requirement: TelegramPlugin uses HTML parse mode
+All outbound message methods in `TelegramPlugin` SHALL use the converter:
+- `sendReply(chatId, text)` — converts text to HTML, sends with `parse_mode: 'HTML'`
+- `sendReplyWithPlaceholder(chatId, text, media, placeholderId)` — converts text, edits placeholder with HTML, sends remaining chunks with HTML
+- Both methods SHALL fall back to plain text on parse errors
+
+#### Scenario: Direct chat reply formatted
+- **WHEN** the AI responds with markdown in a Telegram chat
+- **THEN** the reply SHALL be sent with `parse_mode: 'HTML'` and proper formatting
+
+#### Scenario: Placeholder edit with HTML
+- **WHEN** `sendReplyWithPlaceholder` edits the `...` placeholder message
+- **THEN** `editMessageText` SHALL include `parse_mode: 'HTML'`
+
+### Requirement: Chunking compatibility
+The `splitMessage()` function SHALL work correctly with HTML-formatted text. The system SHALL convert markdown to HTML before chunking, so that HTML tags are not split across chunks.
+
+#### Scenario: Long HTML message chunked
+- **WHEN** a converted HTML message exceeds 4096 characters
+- **THEN** the system SHALL split at newlines or spaces (not inside HTML tags) and send each chunk with `parse_mode: 'HTML'`
diff --git a/src/connectors/telegram/markdown-html.ts b/src/connectors/telegram/markdown-html.ts
new file mode 100644
index 00000000..a30fd965
--- /dev/null
+++ b/src/connectors/telegram/markdown-html.ts
@@ -0,0 +1,71 @@
+/**
+ * Convert standard Markdown to Telegram-compatible HTML.
+ *
+ * Telegram HTML supports: , , , , , .
+ * Headers, tables, and other advanced markdown are converted to approximate
+ * equivalents (bold headers, preformatted tables).
+ */
+
+const PLACEHOLDER = '\x00CB'
+
+interface CodeBlock {
+ placeholder: string
+ html: string
+}
+
+/**
+ * Convert markdown text to Telegram HTML format.
+ * Returns a string safe to use with parse_mode: 'HTML'.
+ */
+export function markdownToTelegramHtml(md: string): string {
+ const blocks: CodeBlock[] = []
+ let blockIdx = 0
+
+ // 1. Extract fenced code blocks before any escaping
+ let text = md.replace(/```(\w*)\n([\s\S]*?)```/g, (_match, _lang, code) => {
+ const ph = `${PLACEHOLDER}${blockIdx++}${PLACEHOLDER}`
+ blocks.push({ placeholder: ph, html: `${escapeHtml(code.trimEnd())}` })
+ return ph
+ })
+
+ // 2. Extract inline code
+ text = text.replace(/`([^`\n]+)`/g, (_match, code) => {
+ const ph = `${PLACEHOLDER}${blockIdx++}${PLACEHOLDER}`
+ blocks.push({ placeholder: ph, html: `${escapeHtml(code)}` })
+ return ph
+ })
+
+ // 3. Escape HTML entities in remaining text
+ text = escapeHtml(text)
+
+ // 4. Headers → bold (## Header → \nHeader\n)
+ text = text.replace(/^#{1,6}\s+(.+)$/gm, '$1')
+
+ // 5. Bold: **text** or __text__
+ text = text.replace(/\*\*(.+?)\*\*/g, '$1')
+ text = text.replace(/__(.+?)__/g, '$1')
+
+ // 6. Italic: *text* or _text_ (not inside bold markers)
+ text = text.replace(/(?$1