Run the Inception — background consolidation agent that clusters vault notes by embedding similarity and produces pattern/synthesis notes. Use PROACTIVELY when the user says "inception", "consolidate", "find patterns", "dream", "run inception", "pattern detection", or invokes /inception. Also use when the user asks about cross-session themes or recurring insights in their vault.

Inception

Runs the Inception consolidation agent against the memento vault. It clusters notes by semantic similarity (HDBSCAN on QMD embeddings) and synthesizes pattern notes via the configured LLM backend.

Process

Determine the run mode from the user's request:
- Default: incremental run (only new notes since last run)
- Dry run: if the user says "dry run", "preview", or "what would inception find" — add --dry-run
- Full: if the user says "full", "all notes", "backfill", or "rescan" — add --full
- Flags can combine: --dry-run --full
Run the Inception script:

python3 ~/.claude/hooks/inception.py --verbose [--dry-run] [--full] [--max-clusters N] 2>&1; echo "EXIT:$?"

Parse the exit code from the last line (EXIT:N) and respond:

Exit	Meaning	Response
0	Success	Parse stderr output for cluster count and notes written. Report to user.
1	Lock held	"Another Inception instance is already running. Try again in a few minutes."
2	Missing deps	"Inception dependencies are missing. Install them with: `pip install numpy hdbscan scikit-learn`"
3	No embeddings	"No embeddings found in QMD. Make sure the vault is indexed: `qmd update -c memento && qmd embed`"
5	Config error	"Configuration error. Check your memento.yml for invalid Inception settings."

For exit 0, summarize:
- How many clusters were found
- How many pattern notes were written (or "dry run — no notes written")
- If notes were written, list their titles
If notes were written, suggest the user review them:
- "You can find the new pattern notes in your vault's notes/ directory"
- "Run /inception --dry-run next time to preview before writing"