Skales v7.0.0 fixes the critical Telegram approval loop and improves agent reliability across the board.
The biggest issue: when using Skales via Telegram in Safe Mode, read-only diagnostic tools like system status checks would trigger an approval prompt every single time. You'd say “yes”, and it would ask again. And again. Endlessly.
The root cause was that 9 internal tools were missing from the safety classification map. Without an explicit classification, they defaulted to requiring manual approval - even though they only read data and have zero side effects.
What changed in v7.0.0
All internal tools are now properly classified. Read-only tools execute automatically, tools with side effects (file writes, emails, commands) still require approval in Safe Mode.
The default for any new or unclassified tool is now “auto-execute” instead of “require approval”. This prevents future tools from silently breaking the approval flow.
Telegram conversations now preserve tool execution history. Previously, after approving a tool, the AI couldn't see that it had already run - so it would try to call it again. Now the full context is maintained.
Google Translate text-to-speech in Telegram was hardcoded to German. It now uses your configured language.
After you approve a tool in Telegram, the AI now produces a natural response (“Done! File created on your Desktop.”) instead of returning raw tool output.
Download v7.0.0 at skales.app. Free for personal use.