fix(plugin/onprem): fix 403 error on profile startup — deadlock + wrong HTTP client + token mismatch

[ArtemSultanov-PG]

Update: Auto token refresh & persistence (2026-07-17)

Added automatic token renewal when expired/invalid, and persistence to files.

What was broken

When a Hermes profile started after being idle (e.g. MacBook off overnight), the stored session token was already expired. The plugin would:

1. Call which calls
2. Get back an expired/invalid token from the API
3. Back off for 60 seconds (_ACTIVATION_RETRY_COOLDOWN)
4. Block ALL subsequent API calls until the cooldown expired — effectively freezing the profile

What was fixed in (_MemantoClient class)

1. New methods for token persistence

• — set profile path for token persistence
• — save session_token to
• — load existing token (future use)

2. updated

• After successful , saves the new token to
• On or , retries activation (auto-refresh)
• No more 60-second freeze after expiration

3. method added

• Tries to activate agent again on token expiration
• Persists the new token to file
• Only runs once per cooldown period

4. updated

• Passes to client via
• So token persistence works automatically

Result

• Tokens are now saved to after every successful activation
• Auto-refresh on expiration — no more profile freeze
• Works correctly after MacBook sleeps, cron refresh, etc.
• Token is persisted across sessions

Files changed

• • Added , , ,
  • Updated with auto-refresh on expired/invalid tokens
  • Updated to pass profile dir to client

Testing

• Verified on profiles: main, server-admin, llm, caraudio, delegate, wrighter, anna-corestone
• All 7 profiles activate correctly after Macbook restart
• Cron job refreshes tokens correctly (script already updated to save tokens)