Tag Curator

~2 min read

Tag Curator

Tags drift over time: #todo vs #todos, #project vs #projects, orphan tags used on one note. This recipe is an agent that audits your taxonomy and proposes a cleanup, applying only after you approve.

Setup

Optional: a lint: schema in index.md (Lettuce’s lint already flags untagged-note and missing-required-tag).
Optional but recommended: a section in AGENTS.md listing canonical tag names + nesting conventions, so the curator unifies toward your vocabulary.

Prompt

@claude Audit my tag taxonomy. Use list_tags then look for:

Near-duplicates (#todo vs #todos, singular vs plural).

Single-use tags that could be merged with something common or removed.

Tags that should be nested (e.g., #web notes that should be #project/web).

Findings already surfaced by lint.

Reply with a table of proposed changes. Don’t apply anything.

Tools the agent will use

Tool	Why
`list_tags`	Get every tag in the vault with note counts.
`notes_by_tag(tag)`	Find every note carrying a near-duplicate, so the merge knows what to touch.
`lint`	Reuse the existing tag-typo / untagged findings rather than re-detecting them.

To apply approved changes:

| patch_section(path, target_type: "frontmatter", target: "tags", op: "replace", content: …) | Rewrite a note’s tag list. | | apply_edits(path, edits) | For inline #tag occurrences in the body. |

Trust + safety

Audit-first: the proposal is a table or a fresh note. Approve before any writes happen.
Each merge is one attributed edit per note. If a merge changes a tag that shouldn’t have moved, the timeline pane reverts it (per-note).
Consider a scoped, write-only-to-tags workflow by pairing this with patch_section‘s frontmatter targeting — the agent can’t accidentally rewrite a note’s body.

Example output

Action	From	To	Notes affected
Merge	`#todos`	`#todo`	4
Nest	`#web`	`#project/web`	7
Remove (single-use)	`#prototype-q3`	—	1

Reply “Apply all” → agent calls patch_section per row, you see seven new entries in the timeline pane, all revertible.