client-workspace

name	client-workspace
description	Scaffold clients/<slug> folder layout. Use when new client folder or workspace layout. Also the shelf step: organize scattered docs by entity and period before extract (see shared/shelf-first.md).

/client-workspace

Purpose

Keep client work isolated, path-addressable, and bookkeeper-shaped.
Organize (shelf) before extract. Doctrine: shared/shelf-first.md.

When to run

• New job / messy folder dump / user pastes many paths
• User: “organize these files”, “set up the client folder”
• Before extract when files are not already under clients/<slug>/source/

If engagement_state.json exists and layout is already standard → only refresh register / add new files.

Default layout (one job)

{workspace}/clients/{client-slug}/
  README.md
  engagement_state.json
  source/
    register.md
    inbox/          # optional unsorted arrivals
    bank/
    sales/
    purchases/
    payroll/
    statutory/
    prior/
    other/
    _pointers.md    # optional: originals not copied
  workpapers/
  outputs/
  ledger/
  queries.md

Config path alternative: ~/.claude/plugins/config/claude-for-accounting/clients/<slug>/

Workflow (shelf-first)

1. Discover — cwd + user-mentioned paths (do not scan whole home).
2. Inventory — kind + period + entity guess (shared/shelf-first.md).
3. Job map — if multi-entity, table + soft-confirm which job now.
4. Scaffold — create layout for active slug.
5. Place — copy (default) or pointer into source/<kind>/; never delete originals.
6. Register — write source/register.md + coverage matrix.
7. Hand off to smart-intake / extract for that shelf only.

Commands (natural language)

• create <name> — slugify, scaffold folders, seed README
• shelf / organize — discover + place + register (this skill’s main path)
• switch <slug> — set active client
• status — active client, FY, stage, open queries
• list — known clients under clients/

Rules

• Never mix two clients’ numbers in one workbook without an explicit multi-entity job.
• Cross-client context is off by default.
• Record the active path so later skills write to the right place.
• Extract/classify/post only from the active job’s shelf (or registered pointers).

Completion

Done when: standard dirs exist; sources are under source/** (or _pointers.md); source/register.md present; active slug clear.