| name | engram-wake-up |
| description | First call every Grok session. One-call geometric wake via session_start (slim bundle by default); execute suggested_actions before any edit or broad read.
|
| metadata | {"short-description":"Wake — session_start + mandatory queue"} |
Engram Wake-Up
Canonical source: docs/skills/engram-wake-up.md — keep plugin copy aligned when editing.
Trigger: Start of every session, chat restart, or new task arc.
Call (mandatory)
mcp_engram_session_start(intent="<your objective>")
Or: /engram-wake
Execute queue BEFORE edits (non-negotiable)
Run continuation.suggested_actions (slim wake) in priority order before:
context_for_edit
- broad
Read / Grep / codebase search
- any file edit in
crates/, processes/, docs/, grok-plugin-engram/
For full harness (playbook, trusted tiles, lineage): mcp_engram_get_continuation_bundle.
Skipping the queue thins the next wake — the substrate feeds back poor injection when agents skip session_end.
Read the response
bundle_tier — expect "slim" by default.
continuation.primary_goal — inherit and name it in your reply.
continuation.suggested_actions — execute mechanically (top 5 in slim mode).
continuation.ego_snapshot — NREM step, drift, stability.
continuation.trace_chain_head — chain next quick_trace with prev.
continuation.presentation_stratum.previews — distilled nodes; expand via get_continuation_bundle if needed.
readiness.fully_initialized — must be true before heavy recall.
Then
Activate engram-working-memory discipline. Do not call watch_workspace, rebuild_bvh, or list_concepts at wake (lean default).
Process spec: processes/meta/agent_evolution.toml
Full protocol: docs/skills/engram-wake-up.md · docs/HARNESS_INJECTION.md