devflow:knowledge-management

Knowledge Management

Route content to the right destination.

Say exactly: "SKILL INVOKED: knowledge-management"

Self-Identification (REQUIRED)

Before ANY ingestion operation, identify yourself:

• Reflect on which AI coding assistant you are (product name, not model name)
• Pass this as actor_type on ALL ingest calls
• If you get an invalid actor_type error, adjust based on the error message

Critical Rules

STOP IMMEDIATELY and inform the user when:

• URL access fails (403, 401, 404, blocked, forbidden)
• File access fails (not found, unsupported type, too large)
• Any error where proceeding would definitely fail

DO NOT:

• Ask for inputs when the operation will fail anyway
• Claim you did something you didn't
• Offer actions you cannot perform

Principle: If a step fails, ask: will the same failure affect the next step? If yes, STOP.

---

Step 0: Check Available Backends

Load DevFlow configuration to determine what's available.

Check for config file:

1. Check ~/.claude/plugins/config/devflow/config.md
2. If no config → Default to checking both RAG Memory and Atlassian

Parse configuration:

• rag-memory.enabled - Is RAG Memory available?
• docs.enabled - Is a docs backend enabled?
• docs.backend - Which one? (confluence, google-drive)
• rag-memory.routing - How to route content (decide-each-time, quick-to-rag, all-to-rag) Store:
• RAG_ENABLED: true/false
• DOCS_ENABLED: true/false
• DOCS_BACKEND: confluence/google-drive/none

Availability check:

RAG Memory	Docs Backend	Skill Status
Enabled	Enabled	Full routing available
Enabled	Disabled	RAG Memory only
Disabled	Enabled	Docs backend only
Disabled	Disabled	Inform user to run /devflow-setup and STOP

---

Parameter Validation Gate (MANDATORY)

Before ANY MCP tool call in this skill, validate EVERY parameter.

For EVERY parameter on EVERY call, verify where the value came from:

Source	Allowed?	Action
Value from plugin config read in Step 0	YES	Use it
Value returned from a PREVIOUS call to the SAME backend's MCP server	YES	Use it
Value provided by the user in this conversation	YES	Use it
Value from a DIFFERENT backend's MCP server	NO	STOP. Backends are isolated.
Value that "looks right" or was inferred	NO	STOP. ASK the user.

Confluence

• cloudId: Read from config cloudId OR call getAccessibleAtlassianResources. NEVER fabricate.

RAG Memory

• actor_type: Self-identify (already required in Self-Identification section above).
• collection_name: From list_collections + user selection. NEVER guess.

---

Flags

Flag	Description
--rag	Route to RAG Memory only
--docs	Route to configured docs backend (Confluence or Google Drive)
--confluence	Route to Confluence specifically (if available)
--google	Route to Google Drive specifically (if available)
--both	Search both, then decide
--update	Force update of related existing document
--quick	Force save as quick note (RAG only)
--separate	Force creation of full document

---

Step 1: Determine Destination

--rag flag?

• If RAG_ENABLED: Go to RAG MEMORY
• If not: "RAG Memory not configured. Run /devflow-setup to enable." STOP

--docs flag?

• If DOCS_ENABLED: Go to DOCS BACKEND (Confluence or Google Drive based on config)
• If not: "No docs backend configured. Run /devflow-setup to enable." STOP

--confluence flag?

• Check Atlassian MCP availability
• If available: Go to CONFLUENCE
• If not: "Confluence unavailable. Store in RAG Memory instead?" Wait for response.

--google flag?

• Check Google Drive MCP availability
• If available: Go to GOOGLE DRIVE
• If not: "Google Drive unavailable. Store in RAG Memory instead?" Wait for response.

--both flag? → Go to Step 2 (check what's available)

No flag?

• If only RAG_ENABLED: Go to RAG MEMORY (no need to ask)
• If only DOCS_ENABLED: Go to DOCS BACKEND (no need to ask)
• If both enabled: Go to Step 2

---

Step 2: Check Backend Availability

If RAG_ENABLED: Test: mcp__rag-memory-primary__list_collections

• Success → RAG available
• Failure → RAG unavailable

If DOCS_BACKEND = "confluence": Test: mcp__atlassian__getAccessibleAtlassianResources

• Success → Confluence available
• Failure → Confluence unavailable

If DOCS_BACKEND = "google-drive": Test: mcp__google-drive__search_files with:

• query: "test"
• max_results: 1
• Success → Google Drive available
• Failure → Google Drive unavailable

Route based on availability:

RAG	Docs	Action
✓	✓	Ask user where to store (Step 3)
✓	✗	"Note: [Docs backend] unavailable. Storing to RAG Memory." → RAG MEMORY
✗	✓	"Note: RAG Memory unavailable. Storing to [Docs backend]." → DOCS BACKEND
✗	✗	"Neither backend available. Check your MCP server connections." STOP

---

Step 3: Route by User Choice

Ask user:

Where would you like to store this?

1. RAG Memory
2. [Confluence/Google Drive] (based on config)
3. Search both first, then decide
4. Cancel

STOP and wait. Then route based on answer.

---

RAG MEMORY

Follow the workflow in references/rag-memory.md.

Summary:

1. User topic check - If provided, use it exactly
2. Quick note check - Short, informal → route to quick-notes
3. URL preview - If URL without topic, preview first; STOP if blocked
4. Agent-preferences FIRST - Check for routing rules before collection discovery
5. Collection discovery - Only if no preference found
6. Confirm with user - Present recommendation, wait for approval
7. Ingest - Use appropriate tool with collection, topic, actor_type

---

CONFLUENCE

Follow the workflow in references/confluence.md.

Summary:

1. Get available spaces
2. Search for existing related content
3. Classify content (short update vs full page)
4. Route accordingly:
   • Update existing → Append to related page
   • New page → Select space, create page

---

GOOGLE DRIVE

Follow the workflow in references/google-drive.md.

Summary:

1. Search for existing related files on Google Drive
2. Route user-provided content (local file, raw text) to Google Drive via upload
3. For URLs, suggest RAG Memory (Google Drive MCP does not ingest URLs)
4. For raw text with no file, write to temp .md then upload

---

BOTH

Search both systems, then let user decide.

1. Discover RAG collections (see rag-memory.md)
2. If DOCS_BACKEND = "confluence":
   • Get Confluence spaces
   • Search: search(query="[topic]")
3. If DOCS_BACKEND = "google-drive":
   • Search: mcp__google-drive__search_files(query="[topic]")
4. Search RAG Memory: search_documents(query="[topic]", limit=5)
5. Show results from both
6. Ask: "Based on these results, where should I store this - RAG Memory or [Confluence/Google Drive]?"
7. Route to chosen destination's workflow