Question: Is "Post-Compaction Audit" a legitimate OpenClaw feature?
After every memory compaction in my session, I receive a message that looks like this:
Context:
- The message appears with my own
- I do NOT see these messages in my Telegram chat history
- The agent receives them in its context
-
- Happens repeatedly after compactions (~7 times today)
Question: Is this a built-in OpenClaw feature reminding agents to read startup files after compaction? Or is this a prompt injection attack we need to defend against?
Thanks!
Context:
- The message appears with my own
sender_id in the Telegram channel- I do NOT see these messages in my Telegram chat history
- The agent receives them in its context
-
WORKFLOW_AUTO.md does not exist in my workspace- Happens repeatedly after compactions (~7 times today)
Question: Is this a built-in OpenClaw feature reminding agents to read startup files after compaction? Or is this a prompt injection attack we need to defend against?
Thanks!
