Load persistent memory from MemCP at the start of every session. Use this skill at the very beginning of a new conversation, before doing any work. Also use when the user says "load memory", "recall context", "what do you remember", or asks about previous sessions. This skill ensures continuity across sessions by loading critical decisions, project context, and stored knowledge.
Load cross-session persistent memory so you have full context before starting any work.
Without loading memory, you start every session from scratch — missing past decisions, known gotchas, user preferences, and ongoing work context. MemCP stores this knowledge persistently. Loading it first prevents re-discovering things the user already told you and avoids repeating past mistakes.
Execute these steps in order. Use parallel tool calls where possible (steps 1-2 can run in parallel).
memcp_recall(importance="critical")
These are non-negotiable rules and decisions — things like coding conventions, architectural constraints, and user preferences that must always be respected.
memcp_recall(project="StochasticMuZero", limit=15)
Recent decisions, findings, TODOs, and experiment results specific to this project. The limit=15 balances coverage with context budget.
memcp_list_contexts()
See if there are any named contexts from previous sessions (e.g., session summaries, analysis results, experiment logs). If any look relevant to the current task, mention them to the user — they can ask you to load specific ones with memcp_get_context(name).
Briefly tell the user what you loaded. Keep it concise — 2-4 bullet points covering:
Example output:
Loaded MemCP memory:
- 3 critical rules (zh-TW convention, config dynamic attrs, build notes)
- 4 project insights (solver supervision design, experiment plan, API usage)
- 1 stored context available:
session-2026-03-08-summary- Open TODO: Phase A infrastructure not yet started