session-continuity

Session Continuity

Use Engram as an active resume layer, not just a memory dump.

The goal is simple:

• start by loading the latest structured handoff
• work normally while keeping important state in Engram
• end with a handoff the next session can use immediately

When to use this

Use this skill when:

• the task spans more than one turn or session
• a second agent may continue the work later
• there are important decisions, blockers, or open loops
• the repo or environment takes time to reconstruct from scratch

Startup flow

At the start of a session:

1. Call resume_context first.
2. If the result is thin, call session_summary or a targeted recall.
3. Treat the returned open_loops, decisions, and recent_work as the default resume packet.

This is better than trying to rebuild context from raw diary entries or a large recall dump.

During work

Keep the handoff current while you work:

• use diary_write when a short progress note will matter later
• use remember for verified facts and implementation state
• use remember_decision for tradeoffs and rationale
• use remember_negative for limitations, unsupported paths, and things that should not be assumed

Engram refreshes the active session handoff automatically after diary writes and important memory writes, so you do not need to manually rebuild the handoff every few minutes.

Before stopping

Near a natural stop point:

1. Call session_handoff.
2. Save it unless you have a reason not to.
3. Check that it captures:
   • current state
   • key decisions
   • recent work
   • open loops

If the next agent should resume the same thread of work, the handoff should make sense without the full chat log.

Suggested pattern

start:
- resume_context

during work:
- diary_write for meaningful progress
- remember / remember_decision / remember_negative as needed

before stopping:
- session_handoff

Quality bar

Good continuity state is:

• specific
• recent
• actionable
• easy to resume from