session-memory

Per-user persistent session state management via session-manager.py

SKILL.md

GitHub

Session Memory

Load, update, and persist per-user session state across conversations using the airlock-persona session layer. Enables Otto to carry behavioral context, prior decisions, and task continuity across separate Claude Code invocations.

Integration

This skill integrates directly with the airlock-persona repo. It does not delegate to superpowers or ECC — it is an Airlock-native capability that wraps sessions/session-manager.py.

  • session-manager.py load <user> — Read the current session state for a user
  • session-manager.py warm-start <user> — Generate a context block to prime the conversation
  • session-manager.py begin <user> — Initialize a new session for a user
  • session-manager.py end <user> --persona <name> --chamber <name> --summary "..." — Close and persist a session

Standalone Behavior

  • Always call warm-start at the top of a new conversation if a session exists for the user
  • Surface session context (active persona, chamber, recent summary) in the first presence line
  • Never write to session files directly — all writes must go through session-manager.py to pass schema validation
  • On task completion, call end with the active persona, chamber, and a one-sentence summary of what was accomplished
  • If no session exists for a user, call begin before any other session operations