Within vs between sessions

Claude's memory works two ways - full within a session, summarised between sessions. Understanding this distinction is what makes multi-session work feel natural instead of broken.

Within a session - full memory

For the duration of one session, Claude remembers everything:

• Every message you typed
• Every file it read
• Every decision made
• Every tool called
• The exact wording you used

Ask "what did we decide about the homepage copy 10 minutes ago?" and Claude quotes it back exactly. This is the context window at work.

Between sessions - summary memory

Once you /wrapup and close the session, the full context is gone. What survives is what /wrapup saved to Supabase:

• A session summary (a paragraph or two)
• Key actions taken
• Decisions made (with reasoning)
• Rules you set or preferences you expressed
• Pending items for next session

That's it. Not every message, not every file, not the exact words. The distilled version.

Why this split exists

Full conversation history = enormous token cost. Saving every word of every session forever would be:

• Prohibitively expensive to load on /startup
• Noisy - 80% of any session is back-and-forth refinement that isn't worth replaying
• Outdated fast - details that mattered in the session often don't matter after

Summary memory gives you continuity without the bloat. Claude remembers the signal, drops the noise.

How to get the best continuity

Write better /wrapup summaries

/wrapup uses what's in the session to build the summary. Noisy sessions produce noisy summaries. Clear sessions produce useful summaries.

At the END of a session, before wrapping up, tell Claude

Before I wrap up - summarise the key decisions we made and the reasoning behind each one. Also list anything I should remember for next session even if it seems obvious.

This gets Claude to generate a high-quality snapshot, which /wrapup then saves.

Use pending items aggressively

Pending items are the breadcrumb for next session. Before /wrapup:

List everything that's still pending for this project. Include blocked items, items I've started but not finished, and anything I decided to postpone. Group by priority.

Claude produces a specific list. /startup next time loads those items - you pick up exactly where you left off.

Don't rely on Claude remembering conversational nuance

Anything important said IN a session - a rule, a decision, a preference - needs to be explicitly saved or it's gone after /wrapup. Tell Claude:

Save this as a rule in my global_rules: [the rule]. Category: feedback.

Or

Add this to the project record for [project-key]: [the fact]. This should persist across sessions.

The ramp-up cost

Between sessions, there's always a brief ramp-up where Claude re-reads relevant files and re-orients itself. /startup minimises this by loading the context deliberately. Without /startup, you end up explaining yourself for the first 5-10 minutes.

Better /wrapup → better memory → less ramp-up. Build the discipline.

Action items

☐ Remember: full memory within session, summary memory between sessions

☐ Before /wrapup, ask Claude to summarise key decisions + pending items

☐ Save important rules explicitly to global_rules or project records - don't assume Claude will remember

☐ Always run /startup at the start of the next session

Next lesson: The day-to-day workflow in practice.

Exercises

1. Review the concepts covered in this lesson: Within vs between sessions.
2. Write down your key takeaway from this lesson.
3. Practice running any commands or prompts mentioned above inside your terminal.