new-workspace

Provision workspace-foundational Workspace

Creates a new generic-purpose workspace. This plugin's commands (/workspace-foundational:setup-workspaces, /workspace-foundational:context-sweep, etc.) are globally available once installed — this skill only provisions the data scaffold (CLAUDE.md, context/, input/, output/, etc.) those commands read from and write to.

Arguments

$ARGUMENTS is parsed as:

• First positional: workspace name (kebab-case, used as directory and GitHub repo name). Required.
• Second positional (optional): target parent path. Defaults to ~/repos/github/my-repos.
• --variant=<name> (optional): which scaffold to copy. Default: generic-workspace. Valid values:
  • generic-workspace
  • context-management
  • report-parsing
  • inventory-analysis
  • web-analytics
  • template-discovery
• --local-only (optional): skip GitHub repo creation and push. Default: create a public GitHub repo and push.
• --private (optional): create the GitHub repo as private. Default: public.

Examples

/workspace-foundational:new-workspace my-notes
/workspace-foundational:new-workspace reports-q2 --variant=report-parsing
/workspace-foundational:new-workspace garage-inventory --variant=inventory-analysis
/workspace-foundational:new-workspace site-analytics --variant=web-analytics --local-only
/workspace-foundational:new-workspace rag-store --variant=context-management

Procedure

1. Parse arguments

Extract workspace name, target parent path, variant, and flags from $ARGUMENTS. If workspace name is missing, ask the user for it before proceeding.

2. Resolve the scaffold path

The bundled scaffold lives at ${CLAUDE_SKILL_DIR}/../../template/<variant>/. Confirm it exists. If the variant isn't one of the six listed above, tell the user which variants are available.

3. Read ambient facts

Read ~/.claude/CLAUDE.md if it exists. Extract OS, locale, timezone, and user identity facts. These will personalise the workspace's CLAUDE.md at step 5.

4. Create the workspace directory

mkdir -p <target-parent>/<workspace-name>
cp -r ${CLAUDE_SKILL_DIR}/../../template/<variant>/. <target-parent>/<workspace-name>/

Do not copy any .claude/ tree. The plugin's primitives are global.

5. Personalise CLAUDE.md

Open the new workspace's CLAUDE.md and:

• Replace any placeholder identity/OS/locale with the facts from step 3.
• Add a short header noting the workspace name and variant.

6. Prompt for workspace-specific facts

Ask the user only for facts the variant actually needs:

• context-management: the Pinecone index + namespace this workspace will maintain.
• report-parsing: the kind of report being parsed (news, academic, legal, etc.) so the triage prompts can tune themselves.
• inventory-analysis: the source system the inventory export came from (Snipe-IT, Airtable, custom CSV, etc.).
• web-analytics: which analytics provider the exports are from (GA4, Plausible, Matomo, etc.).
• generic-workspace / template-discovery: no extra prompts by default.

7. Initialise git and (optionally) publish

cd <target-parent>/<workspace-name>
git init
git add .
git commit -m "Initial workspace from workspace-foundational plugin"

Unless --local-only is set:

gh repo create <workspace-name> --<public|private> --source=. --push

Use --public by default, --private if flag was passed.

8. Print next steps

Tell the user:

• Workspace path and variant.
• Which plugin primitives apply to the chosen variant (e.g. /workspace-foundational:report-triage for report-parsing, /workspace-foundational:inventory-scan for inventory-analysis).
• For template-discovery: suggest running the find-template skill when they describe a new project.
• Reminder that the workspace is data — they can delete/move it freely without losing the plugin's commands.

Notes

• The scaffold path must be resolved via ${CLAUDE_SKILL_DIR}/../../template/ (not ${CLAUDE_PLUGIN_ROOT} — that variable isn't exported in skill bash injection, only in hooks/MCP).
• Never copy .claude/commands/, .claude/agents/, or .claude/skills/ into the new workspace.
• Don't hard-code any personal paths or identifiers here — everything comes from user memory or prompts.