Find your landing page message
Connect a product’s value to a specific audience and write a clear first-screen message.
Probe which emoji reactions are accepted in a specific Telegram chat/message, record an allow/deny list, and optionally remove test reactions afterwards.
---
name: telegram-reaction-prober-265249d375ee
description: Probe which emoji reactions are accepted in a specific Telegram chat/message, record an allow/deny list, and optionally remove test reactions afterwards.
---
# Telegram Reaction Prober
## What this skill is
Telegram reactions are **chat-specific**: a bot can only react with emojis that are enabled for that chat/message. When you try an unsupported emoji, Telegram returns:
- `400 Bad Request: REACTION_INVALID`
This skill is the *prober* (the method + starter datasets), not a global list.
It helps you:
- test a list of candidate emojis against a specific `message_id`
- classify them as **Allowed** vs **Rejected**
- optionally remove successful reactions after testing (so you don't spam the chat)
- write the results into `TOOLS.md` (or another file)
> Important: **Do not publish your private chat’s allow/deny list** as “the answer for everyone”.
> Share the prober + candidate sets; each user still probes their own chat.
## Limits / Reality check
There is no practical way to "test every emoji" (Unicode is enormous, and this Clawdbot Telegram integration does not expose Telegram's enabled-reaction list directly).
So the best approach is:
1) Test a **curated emoji set** you care about (common reactions)
2) Save the whitelist for that chat
3) Re-run when Telegram chat settings change
## How to run (manual)
1) Pick a target Telegram `message_id` (e.g. the most recent message in the chat).
2) Pick a candidate emoji set:
- start small (15–30), or
- use the included 200-emoji starter list:
- `skills/telegram-reaction-prober/assets/emoji200-unicode-frequency-2019.txt`
3) For each emoji:
- call `message` tool with `action=react` and that emoji
- if it succeeds → mark Allowed
- if it fails with `REACTION_INVALID` → mark Rejected
4) Cleanup options:
- **Fast/quiet:** don’t remove; Telegram generally keeps only one reaction per user/bot (so you’re effectively just flipping it)
- **Clean:** remove successful reactions after testing via `remove=true`
5) Write results to `/home/ubuntu/clawd/TOOLS.md` under a heading like:
- `### Telegram reactions (this chat) — tested on message_id XYZ`
## Suggested starter emoji set
Allowed in many chats:
- 👍 ❤️ 🔥 🙏 🎉 🤔 👀
Often rejected (depends on chat):
- ✅ 😂 💡
Good candidates to test next:
- 👎 😅 😭 😮 😍 🤝 👏 🙌 💯 😡 😴 🧠 🧩 ⚠️
## Notes
- Be mindful of rate limits; don’t blast 100+ reactions without delays.
- Keep tests on a single known `message_id` and record it.
---
name: telegram-reaction-prober-265249d375ee
description: Probe which emoji reactions are accepted in a specific Telegram chat/message, record an allow/deny list, and optionally remove test reactions afterwards.
---
# Telegram Reaction Prober
## What this skill is
Telegram reactions are **chat-specific**: a bot can only react with emojis that are enabled for that chat/message. When you try an unsupported emoji, Telegram returns:
- `400 Bad Request: REACTION_INVALID`
This skill is the *prober* (the method + starter datasets), not a global list.
It helps you:
- test a list of candidate emojis against a specific `message_id`
- classify them as **Allowed** vs **Rejected**
- optionally remove successful reactions after testing (so you don't spam the chat)
- write the results into `TOOLS.md` (or another file)
> Important: **Do not publish your private chat’s allow/deny list** as “the answer for everyone”.
> Share the prober + candidate sets; each user still probes their own chat.
## Limits / Reality check
There is no practical way to "test every emoji" (Unicode is enormous, and this Clawdbot Telegram integration does not expose Telegram's enabled-reaction list directly).
So the best approach is:
1) Test a **curated emoji set** you care about (common reactions)
2) Save the whitelist for that chat
3) Re-run when Telegram chat settings change
## How to run (manual)
1) Pick a target Telegram `message_id` (e.g. the most recent message in the chat).
2) Pick a candidate emoji set:
- start small (15–30), or
- use the included 200-emoji starter list:
- `skills/telegram-reaction-prober/assets/emoji200-unicode-frequency-2019.txt`
3) For each emoji:
- call `message` tool with `action=react` and that emoji
- if it succeeds → mark Allowed
- if it fails with `REACTION_INVALID` → mark Rejected
4) Cleanup options:
- **Fast/quiet:** don’t remove; Telegram generally keeps only one reaction per user/bot (so you’re effectively just flipping it)
- **Clean:** remove successful reactions after testing via `remove=true`
5) Write results to `/home/ubuntu/clawd/TOOLS.md` under a heading like:
- `### Telegram reactions (this chat) — tested on message_id XYZ`
## Suggested starter emoji set
Allowed in many chats:
- 👍 ❤️ 🔥 🙏 🎉 🤔 👀
Often rejected (depends on chat):
- ✅ 😂 💡
Good candidates to test next:
- 👎 😅 😭 😮 😍 🤝 👏 🙌 💯 😡 😴 🧠 🧩 ⚠️
## Notes
- Be mindful of rate limits; don’t blast 100+ reactions without delays.
- Keep tests on a single known `message_id` and record it.
Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.