Folder structures fail by ambition: seven levels deep, a taxonomy only its author understands, and within a month everyone files into misc/ or nowhere. The structure that survives is shallow (three levels max), organized by how people look for things (purpose, then project — not file type, not org chart), and honest about human behavior: there's an _inbox for the undecided, a placement rule for every top-level folder, and search is embraced as half the system rather than defeated.
What This Skill Produces
- The structure — top levels with per-folder placement rules ("goes here if…"), three levels deep maximum
- The conventions —
_inboxfor undecided,_archiveper area, the naming rule (see filename-convention) - The one-page guide — the structure explained in ten lines, postable where the team will see it
- The migration-lite plan — new structure now, old mess frozen and drained opportunistically (never the big-bang reorganization that stalls at 30%)
Required Inputs
Ask for these if not provided:
- Who files and who finds — team size, roles, and the honest filing culture (a structure for five diligent people differs from one for forty rushed ones)
- The retrieval questions — the actual "where is the…?" questions of the last month; structure follows retrieval, not theory
- The existing mess — top-level inventory of what exists now, and any folders that genuinely work (survivors get kept, not redesigned)
- Boundaries — what does NOT belong here (personal files, another team's domain, things that live in tools)
Framework: The Structure Rules
- Organize by seek-path, not by file type: people look for "the Acme contract," not "PDFs" — top level mirrors the retrieval questions (Clients / Projects / Team Ops / Reference), never (Documents / Spreadsheets / Images).
- Three levels, then stop:
Clients / Acme / Contracts— anything needing level four becomes a naming problem instead (dates and descriptors in filenames). Depth is where filing compliance dies; every level halves it. - Every folder gets a placement rule: one "goes-here-if" sentence per top-level folder, written on the guide. A folder whose rule can't be written in one sentence is two folders or zero.
_inboxabsorbs the undecided: filing friction comes from placement doubt — the_inboxat root accepts anything instantly, and a weekly ten-minute drain files it properly. Undecided-with-a-home beats misfiled-forever; the underscore keeps it sorted to the top.- Search is half the system: the structure's job is browsability of the current and shared; history goes to
_archive(per area, by year) where search — not browsing — retrieves it. Structures that try to make everything browsable forever collapse under their own history.
Output Format
Folder Structure: [drive/team]The Structure
[Top levels with children to depth 3, each top-level annotated: "→ goes here if …"]
_inbox/ → anything, when unsure — drained weekly
The One-Page Guide
[Ten lines: the placement rules · the naming rule · the _inbox habit · the archive-by-year rule · "when unsure: _inbox, not misc"]
Migration-Lite
[Old root frozen as _pre-2026/ (read-only norm) · new filing starts today · drain rule: whenever you fetch an old file, refile it · no big-bang weekend]
Quality Checks
- Top level mirrors the team's actual retrieval questions
- No path exceeds three levels
- Every top-level folder has a one-sentence placement rule
-
_inboxexists with a named drain cadence and owner - Migration is freeze-and-drain, not big-bang
Anti-Patterns
- Do not build taxonomy for taxonomy's sake — every folder must earn its rule
- Do not organize by file type — nobody seeks "spreadsheets"
- Do not exceed three levels — depth is where compliance goes to die
- Do not attempt the big-bang reorg — it stalls at 30% and leaves two messes
- Do not create
misc/— that's_inboxwithout the drain, i.e., the old problem with a new name