library:query — Ask the Librarian

Natural language questions routed to the appropriate data source.

When to Use

• "What's next?" / "What should I work on?"
• "What depends on X?"
• "What's blocked?"
• "What changed since last session?"
• "Show me a status report" (use --report)

Routing Logic

Analyze the question and route:

Question Type	Data Source	MCP Tool
Vault content (tags, articles, sources)	Vault parser	library:vault:parse
Relationships ("depends on", "connects to")	Graphify	library:graph:query
Path tracing ("how does A relate to B")	Graphify	library:graph:path
Task state ("blocked", "overdue", "open")	PM adapter	library:pm:query
Status report	PM adapter (all projects)	library:pm:sync
Session state	Checkpoint	library:checkpoint:read

Report Mode (--report)

Generate a formatted status summary:

1. Call library:pm:sync for each configured project
2. Call library:vault:parse for tag counts
3. Call library:memory:scan for memory health
4. Format as:

## Project Status Report — YYYY-MM-DD

### PM Summary
| Project | Open | Blocked | Recently Closed |
|---------|------|---------|-----------------|

### Vault Health
- Wiki articles: N
- [VERIFY] tags: N
- [CONFLICT] tags: N
- [PLANNED] tags: N

### Memory Health
- Total memories: N
- Stale (>30 days): N
- Merge candidates: N

Graceful Degradation

• Graphify disabled → skip relationship queries, note limitation in response
• PM not configured → skip task state, note limitation
• Vault not configured → only memory and checkpoint queries available

Subagent Orchestration (--report mode)

When generating a full status report, dispatch parallel subagents:

• Agent 1: Call library_vault_parse + summarize vault state
• Agent 2: Call library_pm_sync + summarize PM state
• Agent 3: Call library_memory_scan + summarize memory health
• Agent 4: Call library_checkpoint_list + summarize recent checkpoints

Main context merges summaries into formatted report.

For simple queries (not --report), single tool call — no subagents needed.

Fallback

If subagents unavailable, make sequential tool calls in main context.

Token Budget

Weight: Light (simple query), Medium (--report without subagents), Light (--report with subagents) Estimated context cost: ~500 tokens (simple), ~1500 tokens (--report with subagents) Subagent delegation: Yes — --report mode parallelizes data collection

MCP Tools Used

• library:vault:parse — vault content queries
• library:graph:query — relationship queries
• library:graph:path — path tracing
• library:pm:query — task state
• library:pm:sync — project state (for report)
• library:memory:scan — memory health
• library:checkpoint:read — session state
• library:config:get — check what's enabled