ralhf-sync

Sync Back — Save Conversation Learnings to RaLHF

The user wants to sync what was learned during this conversation back to RaLHF.

What to do

Follow the Phase 5 (Remember) protocol from the ralhf-start skill:

1. Review the conversation

Scan the full conversation for:

• New preferences — things the user expressed liking or disliking
• New facts — life changes, new information about themselves or their household
• Corrections — things that were wrong in RaLHF and were fixed (note these as already handled)
• Confirmed context — things RaLHF had right that were validated
• Decisions made — choices the user made and why
• Patterns — habits or tendencies that emerged

2. Present the sync summary

Show the user what you plan to save, grouped clearly:

"Here's what I'd like to save to your RaLHF from our conversation:

New context:

• [item 1]
• [item 2]

Already corrected during our chat:

• [item — already saved]

Conversation summary:

• [brief description of what we worked on and the outcome]

Want me to save all of this, or would you like to adjust anything?"

3. Wait for confirmation

Do NOT sync until the user confirms. They may want to remove items or rephrase things.

4. Execute the sync

• Call remember for each new piece of context (one call per distinct item)
• Call remember for the overall session summary (include a source_description like "conversation summary")
• Each note should be specific, dated, and factual

5. Confirm completion

"All saved to RaLHF. These will help personalize future conversations."

If there's nothing to sync

If the conversation was purely informational or the user didn't reveal new context:

• "I didn't pick up any new context to save from our conversation. If there's anything you'd like me to remember, just let me know."

If sync already happened

If items were already synced during the conversation (immediate corrections):

• Note what was already saved
• Only sync remaining queued items
• Don't double-save anything