Request early access to Session Memory

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I run 89+ Claude Code sessions for complex multi-dossier work spanning weeks. Cross-session continuity is critical — a strategy decision on Monday must be available Thursday when I return to casework.

Currently I rely on manual /eod discipline, MEMORY.md files, and autoMemoryEnabled. This works but is fragile — if I forget /eod, context is lost. /compact during long analysis sessions is lossy and risky.

My account (tengu_session_memory: false in GrowthBook) is on first-party Anthropic API (not Bedrock/Vertex/Foundry), so I'm technically eligible per the SX() gate — the flag just hasn't been flipped for my account.

Proposed Solution

Toggle tengu_session_memory: true for my account. I understand this may be in staged rollout — I'm requesting early/priority access given my high-usage, high-retention profile. I'm happy to provide feedback on the feature in a professional (non-dev) context.

Alternative Solutions

• autoMemoryEnabled (already active) — reads/writes MEMORY.md directory, partial solution
• autoDreamEnabled (already active) — overnight consolidation, good but no in-session extraction
• Manual /eod + markdown MEMORY.md — current workflow, requires discipline
• Third-party tools — add complexity, prefer native solution

Priority

Critical - Blocking my work

Feature Category

Other

Use Case Example

Monday: 2-hour session analyzing case strategy on a complex legal matter. Key decision made.

Tuesday: Unrelated session on a separate matter.

Wednesday: No session.

Thursday: Return to Monday's case. Session start automatically injects Monday's analysis — Claude picks up where we left off. No manual briefing, no /eod dependency, zero friction.

Additional Context

Also interested in tengu_sm_compact (lossless /compact using pre-written session summaries). Long sessions needing compaction are currently risky — compact can drop critical reasoning chains.