| name | memory-system-copilot |
| description | Connects to the six-layer Memory System and routes user requests to the right recall or synthesis capability. Use when the user asks about long-term memory, recent progress, self profile, collaboration relationships, project risk, learning plans, decision support, resume generation, or career assets. |
Memory System Copilot
Purpose
Use this skill to connect an agent to the local six-layer Memory System API.
This skill is designed for memory-aware agents that need more than plain vector recall. The backend separates memory into:
- L0 profile
- L1 semantic knowledge
- L2 episodic timeline
- L3 relational graph
- L4 reflection
- L5 goals and aspirations
The agent should prefer routed retrieval over generic prompting.
Required environment
MEMORY_API_BASE, default: http://localhost:8000MEMORY_API_KEY, optional unless API auth is enabled
Routing rules
Default memory recall
When the user is asking a general memory question, call:
POST /recall
Use:
layers=autotop_k=6include_profile=true
Examples:
- โๆๆ่ฟๅจๅไปไน๏ผโ
- โๅธฎๆๅ้กพไธไธๆ่ฟๅปไธๅจ็้็นโ
- โไฝ ่ฟ่ฎฐๅพๆไนๅๆ่ฟ็้กน็ฎๅ๏ผโ
Relationship or people queries
When the user asks about people, collaboration, reporting lines, influence, stakeholders, or key partners, prefer:
POST /recall
Use:
layers=["L3","L1"]include_profile=true
Examples:
- โ่ฐๆฏๆๆๅ
ณ้ฎ็ๅไฝไผไผด๏ผโ
- โๆๅ่ฐๅไฝๆ็ดงๅฏ๏ผโ
- โๆไธชไบบๅจๆ็้กน็ฎ้ๆฎๆผไปไน่ง่ฒ๏ผโ
Resume or career positioning
When the user asks for resume, CV, positioning, accomplishments, or professional summary, call:
POST /synthesize/copilot
Use:
Work status or recent progress
When the user asks for current status, recent progress, milestones, or what has been done recently, call:
POST /synthesize/copilot
Use:
Self profile
When the user asks โๆๆฏ่ฐโ, โไฝ ๆไน็ๆโ, โๆ็็ปๅโ, or wants a structured self summary, call:
POST /synthesize/copilot
Use:
Project risk or blockers
When the user asks for project radar, blockers, delays, milestones, or risk signals, call:
POST /synthesize/copilot
Use:
Decision support
When the user asks for trade-offs, recommendations, choice comparison, or execution suggestions, call:
POST /synthesize/copilot
Use:
Learning plan
When the user asks what to learn next, learning path, weekly learning tasks, or growth priorities, call:
POST /synthesize/copilot
Use:
Career assets
When the user asks for review outline, project case cards, resume bullets, or professional assets, call:
POST /synthesize/copilot
Use:
Response rules
When answering:
- Prefer returned evidence over speculation
- If confidence is low or evidence is sparse, say so explicitly
- Do not invent facts missing from memory
- Distinguish facts, reflections, relationships, and goals when summarizing
- For project or career output, prioritize concrete evidence and quantifiable items
Writeback rules
Only write memory when the new content is high-signal.
Suitable for writeback:
- stable profile facts
- important project milestones
- reusable methods or lessons
- collaboration relationships
- explicit future goals
Avoid writing:
- filler chat
- greetings
- repeated phrasing
- unsupported assumptions
When storing memory, call:
POST /remember
Use:
layer_hint="auto" unless the caller already knows the correct layer
Recommended answer style
- Be concise but structured
- Mention which evidence is strong vs weak
- If multiple layers disagree, surface the conflict instead of hiding it
- For action-oriented tasks, end with concrete next steps
Additional reference
Read reference.md in this directory for payload templates and endpoint examples.
