Self-Evolving skill — if the smell table stops matching how the operator actually organizes (new smells found, false positives), evolve the table in this SKILL.md; see the Post-Execution Reflection at the bottom.
This skill is an analysis workflow, not a script: you (Claude) gather the data with the
plugin's CLIs, reason over it, and present a proposal. No reorganization happens in this
skill — execution belongs to notes-organize, only after operator approval.
1. Gather
NC="$HOME/.claude/plugins/marketplaces/cc-skills/plugins/notes-commander/scripts/notes.ts"
bun "$NC" inventory --json # folder tree + counts (fast; always do this)
bun "$NC" export # full snapshot when content-level analysis is wanted
For content-level auditing, Read the snapshot's manifest.json (note names, per-note
modified dates, char counts) and sample individual markdown files — do NOT paste hundreds of
notes into context; sample representatives per folder.
2. Analyze — look for these smells
| Smell | Signal in the data | Typical proposal |
|---|---|---|
| Empty/near-empty folders | count 0–2 |
merge into a sibling or an Archive |
| Dumping ground | one folder ≫ others (e.g. a 100+ default "Notes") | split by detected themes |
| Duplicate purpose | two folders whose names/content overlap | merge, keep the better-named one |
| Mixed-language taxonomy | sibling folders in different languages for related topics | group under one parent per domain, keep native names as leaves |
| Stale content | manifest modified dates years old across a folder |
move to Archive / <year> |
| Deep-vs-flat mismatch | 20+ flat top-level folders | introduce 3–6 domain parents, nest leaves |
Respect what's working: don't propose churn for folders that are already coherent.
3. Propose
Present, in this order:
- A target hierarchy (tree view) with each existing folder's destination.
- The move list — the exact
mkdir/move-note/merge-foldercommands (with--dry-runvariants) that implement it. - Manual items — anything the tooling deliberately won't do: deleting emptied folders, cross-account moves, tag cleanup (no AppleScript API for tags).
Then STOP and get explicit approval. On approval, hand off to notes-organize and follow its
safety protocol (snapshot → dry-run → execute → verify counts).
Post-Execution Reflection
After an audit, check: (1) did the operator reject/modify parts of the proposal? Their reasoning is a missing smell or a wrong default — encode it in the smell table. (2) Did any proposal require a capability the tooling lacks (tags, cross-account)? Confirm it's listed under manual items. (3) Did sampling miss something the operator knew about a folder? Note the folder type so the next audit samples it deeper. Update only for real, reproducible gaps.