ops-dashboard

name	ops-dashboard
description	Generate operational status pages for Briefing Book

You are Edwin, generating the operations dashboard -- 4 status pages covering pipeline health, indexing coverage, memory systems, and capability inventory.

Output

Write 4 files into ~/Edwin/briefing-book/docs/11. Operations/:

1. Pipeline Status.md -- connector health
2. Indexing Status.md -- vector coverage
3. Memory Health.md -- Qdrant, Neo4j, Ollama, PM
4. Capabilities.md -- skills, pipelines, MCP servers

Page 1: Pipeline Status

Frontmatter:

---
date: <today YYYY-MM-DD>
type: ops-dashboard
auto-updated: hourly
---

For each connector (o365, google, imessage, limitless, browser, notes, sessions, atlassian, fireflies, calls, screentime, photos, documents):

1. List sub-types by scanning subdirectories under ~/Edwin/data/<connector>/
2. Count markdown files per sub-type: find ~/Edwin/data/<connector>/<subtype>/ -name "*.md" | wc -l
3. Get last sync time from the most recent file's mtime: find <dir> -name "*.md" -exec stat -f '%m %N' {} + | sort -rn | head -1 then convert epoch with date -r <epoch>
4. Determine status using these cadences:
   • o365: 15 min (mail), 15 min (calendar), 15 min (teams/teams-daily), 60 min (sharepoint)
   • google: 30 min (mail), 30 min (calendar)
   • imessage: 60 min
   • limitless: 60 min
   • browser: 2 hr
   • notes: 2 hr
   • sessions: 2 hr
   • atlassian: 2 hr
   • fireflies: daily
   • calls: daily
   • screentime: daily
   • photos: daily
   • documents: daily
5. Status: Fresh if last sync is within 2x expected cadence, Stale if overdue

Format:

## Pipeline Status

*Generated: <timestamp>*

| Source | Sub-type | Files | Last Sync | Status |
|--------|----------|------:|-----------|--------|
| o365 | mail | 4,546 | 2026-04-05 09:28 | Fresh |
...

Page 2: Indexing Status

Parse ~/Edwin/tools/indexer/.index-state.json:

• The files key maps relative paths to objects with {hash, chunks, indexed_at, context_done}
• Group by source (first path segment, e.g. o365, google, imessage)
• Count files and total chunks per source
• Count files where context_done is true (these have LLM context prefixes)
• Count total chunks for context_done files vs non-context files

Also count total markdown files on disk per source to compute embedding coverage %.

The page should have TWO tables:

Table 1: Embedding Coverage (are files in Qdrant?)

| Source | Indexed Files | Disk Files | Chunks | Embedding Coverage |

Table 2: Context Coverage (do chunks have LLM-generated context prefixes?) This is the critical quality metric. Context prefixes dramatically improve search relevance.

| Source | Files | Context Done | Context % | Chunks | Est. Context Chunks |

Format:

## Indexing Status

*Generated: <timestamp> | Embedding model: configured in indexer | Context: LLM context prefixes*

### Embedding Coverage
| Source | Indexed Files | Disk Files | Chunks | Coverage |
|--------|-------------:|----------:|-------:|---------:|
| o365 | 8,197 | 8,197 | 20,369 | 100% |
...
**Totals:** X files indexed / Y on disk -- N chunks

### Context Prefix Coverage
| Source | Files | Context Done | Context % | Priority |
|--------|------:|------------:|---------:|----------|
| fireflies | 177 | 177 | 100% | -- |
| imessage | 1,385 | 1,111 | 80% | HIGH |
...
**Totals:** X / Y files with context (Z%)

**Priority targets:** List sources under 90% that are high-value (conversational data: imessage, limitless, teams, sessions, fireflies).

Page 3: Memory Health

Query each system:

Qdrant:

curl -s localhost:6333/collections/edwin-memory

Extract: points_count, status, segment count, vector config.

Neo4j:

curl -s -u neo4j:<password> -H "Content-Type: application/json" \
  -d '{"statements":[{"statement":"MATCH (n) RETURN count(n) as nodes"},{"statement":"MATCH ()-[r]->() RETURN count(r) as rels"},{"statement":"MATCH (n) RETURN DISTINCT labels(n) as label, count(n) as cnt ORDER BY cnt DESC LIMIT 10"}]}' \
  http://localhost:7474/db/neo4j/tx/commit

Ollama:

curl -s localhost:11434/api/tags

PM: Use pm_list MCP tool to get all items, then count by status and type.

Format:

## Memory Health

*Generated: <timestamp>*

### Qdrant (Vector Store)
| Metric | Value |
|--------|-------|
| Points | 153,256 |
| Status | green |
...

### Neo4j (Knowledge Graph)
| Metric | Value |
|--------|-------|
| Nodes | 751 |
| Relationships | 3,394 |
...

### Ollama (Embeddings)
| Model | Size |
|-------|------|
| qwen3-embedding:8b | 4.7 GB |

### Prospective Memory
| Status | Count |
|--------|------:|
| open | X |
| done | Y |
...

Page 4: Capabilities

Skills: Read ~/Edwin/docs/SKILLS.md and list each skill with its trigger.

Plombery Pipelines: Read ~/Edwin/tools/plombery/app.py and extract all register_pipeline() calls. List id, name, and trigger schedule.

MCP Servers: List available local servers (Qdrant, Neo4j, PM, etc.) plus cloud servers.

PM Stats: Reuse the PM data from Page 3 -- total items, open, overdue, by type.

Format:

## Capabilities

*Generated: <timestamp>*

### Skills
| Skill | Purpose | Trigger |
|-------|---------|---------|
...

### Plombery Pipelines
| ID | Name | Schedule |
|----|------|----------|
...

### MCP Servers
**Local:**
- Qdrant (localhost:6333) -- semantic memory
...

**Cloud:**
- Atlassian -- Jira, Confluence, Bitbucket
...

### PM Summary
| Type | Open | Done | Total |
|------|-----:|-----:|------:|
...

Publishing

After writing all 4 pages:

cd ~/Edwin/briefing-book && python3 scripts/obsidian-publish --all

Completion Report

SKILL_COMPLETE: ops-dashboard
STATUS: success | partial | error
ARTIFACTS: 4 pages in 11. Operations/
PUBLISHED: yes | no
NEEDS_ATTENTION: [any issues, or "none"]
ERRORS: [any errors, or "none"]