Telegram bridge cannot send files/documents even with --accept-all

[edwardcodelab]

Describe the bug

Cheetahclaws send [approve][reject] which I can't click.

Even after setting accept-all via , the Telegram bridge cannot send files back to the user.

• Text messages, slash commands, voice notes, and photos to cheetahclaws work fine.
• Any time the AI tries to send a file no file appears in Telegram.
• The AI keeps saying “file sent” or shows the file content in the web UI, but nothing arrives in the Telegram chat.

This makes the Telegram bridge unusable for any workflow that involves file output.

Docker:
volumes:
- ./data:/data
- /path/to/workspace:/workspace:rw

[chauncygu]

Thanks for your great feedback.

We already fixed in the latest version, both halves of this issue are addressed:

1. File round-trip. --accept-all was a red herring; bridges/telegram.py simply lacked the upload path. Added:

• _tg_send_document — multipart/form-data upload, 49 MB cap, six explicit failure modes reported back to chat (missing / stat fail / empty / oversize / network / API ok:false).
• Inbound document handler — saves uploads to /workspace (Docker) or tempfile.gettempdir() with sanitized filenames, then prompts the agent with the saved path.
• !sendfile — explicit on-demand send.
• Auto-send on Write — when the model writes a file, it's mailed to the chat as a document. FIFO-paired with the in-flight file_path, skipped on Error: / Denied: results, de-duplicated per turn.

2. Clickable [approve][reject]. The text-bracket pseudo-buttons are now real Telegram inline_keyboard buttons (✅ Approve / ❌ Reject / ✅✅ Accept all). The bridge:

• Widens allowed_updates to ["message", "callback_query"].
• Validates each click's chat_id (auth) and a per-prompt prompt_id baked into callback_data (so stale clicks on older prompts don't bleed in).
• Acknowledges via answerCallbackQuery to dismiss the spinner.
• Edits the original message to append ✓ Selected: y for visible scroll-back.
• Falls back gracefully: Markdown failure → no-parse_mode keyboard → plain text without keyboard. The [y/N/a] text hint stays in the prompt body for fallback paths.

Docs: new "Permission prompts" + "File support" sections in docs/guides/bridges.md

Slack / WeChat / terminal are unaffected — they ignore the new options= parameter on ask_input_interactive and keep using the existing text path. PR coming shortly.

[chauncygu]

All fixed, please use the latest version.

Thanks again for your great feedback.

Best,