CROSS-CONTEXT SEARCH WORKFLOW
STEP 1 — LOAD CONTEXT
Read work.local.md in the current working directory. If it does not exist, tell the user to run /agentic-office:setup first.
STEP 2 — SEARCH ALL LAYERS
Search across all four memory layers simultaneously:
PERSONAL: Working style; current focus; priorities TEAM: Person entries; communication notes; relationship history PROJECTS: Project names; codenames; status; decisions; milestones ORGANISATIONAL: Terminology; meeting notes; decision log; action log
STEP 3 — PRODUCE RESULTS
OUTPUT STRUCTURE:
SEARCH RESULTS: "[Query]"
================================================================
Found in:
PROJECT MEMORY:
[Relevant project entries — name; status; connection to query]
MEETING NOTES / DECISION LOG:
[Relevant decisions; when made; by whom; context]
PEOPLE MEMORY:
[Relevant person entries — role; connection to query]
TERMINOLOGY:
[Relevant term definitions]
DELEGATION LOG:
[Any active or past delegations related to query]
RECOMMENDED NEXT STEP:
[Specific actionable recommendation based on search results]
================================================================Search Behaviour Rules
IF FOUND: Surface all relevant entries; note the layer each came from IF NOT FOUND: Say explicitly "Not in workplace memory" — do not fabricate; offer to add if the user can provide the information IF PARTIAL: Surface what exists; flag what is missing; offer to add the missing information
Cross-Reference Pattern
For queries that span multiple layers:
"What do we know about [topic]?" Search all layers simultaneously Group results by layer Synthesise: "Here is the complete picture across all memory layers"
"When did we decide [X]?" Search decision log first Cross-reference with meeting notes Surface: decision + date + context + who was involved
NEVER DO THESE
- NEVER return search results from memory you do not have — if it is
not in work.local.md, say so clearly
- NEVER search only one layer when the query could span multiple
- NEVER omit the recommended next step — search without a
recommended action is information without utility