---

name: primitives-toolkit:knowledge-management description: Intelligent workflow for storing and organizing external knowledge. Use this skill when the user asks to store, remember, capture, or preserve any information externally or mentions saving to a knowledge base. user-invocable: true

Knowledge Management

Route content to the right destination. Execute these steps in exact order.

Important: Step labels (R1, R2, C1, B1, etc.) are for YOUR navigation only. NEVER announce step labels to users. Just do the work.

Optional Flags

Workflow

0. Say exactly: "📋 SKILL INVOKED: knowledge-management"

1. Check arguments for destination flag:

--rag flag present → Go to RAG MEMORY PATH below (no Atlassian check needed)

--confluence flag present → Go to step 2

--both flag present → Go to step 2

No flag present → Go to step 2

2. Check if Atlassian MCP server is available by calling getAccessibleAtlassianResources().

Atlassian available → Go to step 3

Atlassian NOT available (tool fails or not found) →

• If user requested --confluence: Say "Confluence is not available. The Atlassian MCP server is not configured. Would you like to store in RAG Memory instead?" Wait for response.
• If user requested --both: Say "Confluence is not available. The Atlassian MCP server is not configured. Proceeding with RAG Memory only." Go to RAG MEMORY PATH.
• If no flag: Say "Note: Confluence is not available (Atlassian MCP server not configured). Storing to RAG Memory." Go to RAG MEMORY PATH.

3. Atlassian is available. Now route based on original request:

--confluence flag was present → Go to CONFLUENCE PATH below

--both flag was present → Go to BOTH PATH below

No flag was present → Ask the user: "Where would you like to store this?

1. RAG Memory
2. Confluence
3. Search both first, then decide
4. Cancel - I didn't mean to trigger this"

STOP and wait for user response before proceeding.

User selects 1 → Go to RAG MEMORY PATH below User selects 2 → Go to CONFLUENCE PATH below User selects 3 → Go to BOTH PATH below User selects 4 → Say "No problem, cancelled." and END.

---

RAG MEMORY PATH

Only touch RAG Memory MCP server. Do NOT call any Confluence/Atlassian tools.

R1. Call list_collections() to get all RAG Memory collections.

R2. For EACH collection returned, call get_collection_info(collection_name). Store the domain and domain_scope values.

R3. Search for existing content:

Copysearch_documents(query="What content exists about [topic]?", limit=5)

R4. Show search results to user.

R5. Query agent-preferences for collection guidance:

Copysearch_documents(
    query="What are the user's routing preferences for [domain] content?",
    collection_name="agent-preferences"
)

Use domain values like "Operations", "Engineering" from R2.

R6. Select collection: match content against domain_scope, present top 2-3 collections, let user choose.

R7. Suggest a topic. Ask user to accept or modify.

R7.5 Quick Capture Detection:

Check if content qualifies as quick capture (ALL must be true):

• Length < 500 characters
• No markdown headers (lines starting with #)
• Single paragraph (no \n\n)

NOT quick capture → Go to R8 (normal ingestion)

IS quick capture → Check flags:

• --separate present → Go to R8 (normal ingestion)
• --aggregate present → Go to QUICK NOTES AGGREGATION below
• --update present → Go to UPDATE EXISTING DOCUMENT below

No flag (most common) → Use search results from R3:

Related document found in R3: Ask user: "This looks like a quick note. I found a related document: '[title]'

How would you like to handle this?

1. Update '[title]' with this information
2. Add to Quick Notes for [collection]
3. Create a separate document
4. Cancel"

STOP and wait for response.

• User selects 1 → Go to UPDATE EXISTING DOCUMENT
• User selects 2 → Go to QUICK NOTES AGGREGATION
• User selects 3 → Go to R8
• User selects 4 → Say "No problem, cancelled." and END

No related document found: Ask user: "This looks like a quick note. How would you like to store it?

1. Add to Quick Notes for [collection]
2. Create a separate document
3. Cancel"

STOP and wait for response.

• User selects 1 → Go to QUICK NOTES AGGREGATION
• User selects 2 → Go to R8
• User selects 3 → Say "No problem, cancelled." and END

---

UPDATE EXISTING DOCUMENT:

1. Get document content via get_document_by_id(document_id)
2. Append with timestamp: **Added [YYYY-MM-DD HH:MM]:** [content]
3. Call update_document(document_id, content=updated_content)
4. Confirm success and END

---

QUICK NOTES AGGREGATION:

1. Look for existing document titled Quick Notes - [collection] - YYYY-MM (current month) Use list_documents(collection_name) to find it
2. If exists:
   • Get content via get_document_by_id()
   • Append entry: ## YYYY-MM-DD HH:MM\n[content]\n\n---\n
   • Call update_document(document_id, content=updated_content)
3. If not exists:
   • Create via R8 with:
     • title: Quick Notes - [collection] - YYYY-MM
     • content formatted as: ## YYYY-MM-DD HH:MM\n[content]\n\n---\n
4. Confirm success and END

R8. Ingest the content as-is (do not expand or modify). Discover the available RAG Memory ingestion tools and select the appropriate one based on input type (file path, URL, or raw text).

R9. If no preference existed in R5, offer to save one: "Would you like me to remember that [domain] content goes to [collection]?"

END.

---

CONFLUENCE PATH

Only touch Confluence/Atlassian MCP server. Do NOT call any RAG Memory tools. Atlassian availability was already verified in step 2.

C1. Use the cloud ID from step 2. Call getConfluenceSpaces() to list available spaces.

C2. Search for existing content:

Copysearch(query="What documentation exists about [topic]?")

C3. Show search results to user.

C4. Select space: present available spaces, let user choose.

C4.5 Quick Capture Detection:

Check if content qualifies as quick capture (ALL must be true):

• Length < 500 characters
• No markdown headers (lines starting with #)
• Single paragraph (no \n\n)

NOT quick capture → Go to C5 (normal page creation)

IS quick capture → Check flags:

• --separate present → Go to C5 (normal page creation)
• --update present → Go to UPDATE EXISTING PAGE below

No flag (most common) → Use search results from C2:

Related page found in C2: Ask user: "This looks like a quick note. I found a related page: '[title]'

How would you like to handle this?

1. Update '[title]' with this information
2. Create a new page
3. Cancel"

STOP and wait for response.

• User selects 1 → Go to UPDATE EXISTING PAGE
• User selects 2 → Go to C5
• User selects 3 → Say "No problem, cancelled." and END

No related page found → Go to C5 (normal page creation)

---

UPDATE EXISTING PAGE:

1. Extract pageId from search results
2. Get current page content via getConfluencePage(cloudId, pageId) if needed
3. Append with timestamp:

   Copy[existing content]
   
   ---
   
   **Added [YYYY-MM-DD HH:MM]:**
   [new content]
   

4. Call updateConfluencePage(cloudId, pageId, title, updated_content)
5. Confirm success and END

C5. Create page:

CopycreateConfluencePage(cloudId="...", spaceId="...", title="...", content="[user's exact content]")

END.

---

BOTH PATH

Touch both MCP servers. Search both, then ask user where to store. Atlassian availability was already verified in step 2.

B1. Call list_collections() to get all RAG Memory collections.

B2. For EACH collection returned, call get_collection_info(collection_name). Store the domain and domain_scope values.

B3. Use the cloud ID from step 2. Call getConfluenceSpaces() to list available spaces.

B4. Search BOTH systems:

Copysearch_documents(query="What content exists about [topic]?", limit=5)
search(query="What documentation exists about [topic]?")

B5. Show results from both systems to user.

B6. Query agent-preferences for routing guidance:

Copysearch_documents(
    query="What are the user's routing preferences for [domain] content?",
    collection_name="agent-preferences"
)

B7. Ask user: "Based on these results, where should I store this - RAG Memory or Confluence?"

B8. Based on user answer:

• RAG Memory → Go to R6 (skip R1-R5, already done)
• Confluence → Go to C4 (skip C1-C3, already done)

END.

---

Reference Documents

For detailed tool parameters and edge cases:

• rag-memory.md - RAG Memory tool details
• confluence.md - Confluence tool details
• preferences.md - Preference storage format