new-workspace

name	new-workspace
description	Provision a new research-space workspace on disk. Use when the user wants to start a new research project (deep-research, technical, osint, georeaction, stack, ecosystem, competitor, purchasing, general-research-workspace, or obsidian-vault). Accepts a workspace name and optional variant. Scaffolds the workspace, personalises CLAUDE.md from the user's global memory, creates a GitHub repo (public by default), and auto-registers public workspaces into the Open-Research-Workspaces-Index.
disable-model-invocation	true
allowed-tools	Bash(mkdir *), Bash(cp *), Bash(cat *), Bash(git init *), Bash(git add *), Bash(git commit *), Bash(gh repo create *), Bash(gh auth status), Bash(git push *), Bash(gh repo clone *), Bash(git clone *), Bash(git -C *), Bash(rm -rf /tmp/*), Read, Edit

Provision Research-Space Workspace

Creates a new workspace for open-ended research. This plugin's primitives (/research-space:source-log, /research-space:summarize-sources, /research-space:deep-dive, etc.) are globally available once installed — this skill only provisions the data scaffold (CLAUDE.md, context/, sources/, outputs/, notes/, plus variant-specific folders) that those commands read from and write to.

Arguments

$ARGUMENTS is parsed as:

• First positional: workspace name (Train-Case preferred, kebab-case accepted). Required.
• Second positional (optional): target parent path. Defaults to the topic group under ~/repos/github/ that most specifically fits the subject — see ~/repos/github/README.md for the group list. (~/repos/github/my-repos/ is no longer the default; it holds only blog repos now.)
• --variant=<variant> (optional): which scaffold to copy. Default: deep-research. Supported:
  • deep-research — general-purpose investigation around one central question
  • technical — version-sensitive technical research (APIs, libraries, protocols, hardware)
  • osint — open-source intelligence with provenance and chain-of-custody
  • georeaction — comparative regional reactions to an event or policy
  • stack — tool/stack evaluation with criteria and scoring
  • ecosystem — actor/landscape mapping
  • competitor — competitor intelligence
  • purchasing — spec → candidates → shortlist → decision for a purchase
  • general-research-workspace — openly-logged Q&A research space (user asks, Claude writes longform responses, pairs consolidated into PDFs)
  • obsidian-vault — the standard research loop scaffolded as a working Obsidian vault (committed .obsidian/ config, frontmatter schema, wikilinks, note templates, canvas). Pick this when the user says "Obsidian", "vault", "PKM", or wants to browse the research by graph and backlinks rather than by folder
• --local-only (optional): skip GitHub repo creation and push. Default: create a GitHub repo and push.
• --private (optional): create the GitHub repo as private. Default: public. Private workspaces are not auto-registered into the public index.
• --no-index (optional): even for a public workspace, skip the auto-registration into Open-Research-Workspaces-Index.

Examples

/research-space:new-workspace Iran-Drone-Program --variant=osint
/research-space:new-workspace Acme-Competitor --variant=competitor
/research-space:new-workspace Local-AI-Stack --variant=stack --local-only
/research-space:new-workspace LLM-Ecosystem-2026 --variant=ecosystem
/research-space:new-workspace Framework-Laptop-Purchase --variant=purchasing --private
/research-space:new-workspace State-Of-Claude-Context-0426 --variant=general-research-workspace
/research-space:new-workspace Agent-Memory-Landscape --variant=obsidian-vault

Procedure

1. Parse arguments

Extract workspace name, target parent path, variant, and flags from $ARGUMENTS. If workspace name is missing, ask the user before proceeding. If variant is not one of the supported ten, list the supported ones and stop.

2. Resolve the scaffold path

The bundled scaffold lives at ${CLAUDE_SKILL_DIR}/../../template/<variant>/. Confirm it exists before copying.

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.

The trailing /. matters: it carries dotfiles. For obsidian-vault that is the whole point — without .obsidian/ and .gitignore the result is a folder of markdown, not a vault. After copying that variant, verify:

test -f <workspace>/.obsidian/app.json && test -f <workspace>/.gitignore || echo "dotfiles missing — re-copy with the trailing /."

5. Personalise CLAUDE.md

Open the new workspace's CLAUDE.md and:

• Replace placeholder identity/locale with facts from step 3.
• Insert a header with workspace name, variant, and date.
• Fill in any variant-specific placeholder (see step 6).

6. Prompt for workspace-specific facts

Ask only for facts the plugin can't infer:

• deep-research / technical: the central research question → <RESEARCH_QUESTION>.
• osint: the investigation subject and (optionally) the geographic scope.
• georeaction: the event or policy and the regions being compared.
• stack: the technology area under review → <STACK_AREA>.
• ecosystem: the ecosystem being mapped (geography or sector).
• competitor: the competitor company name and your own company for framing.
• purchasing: the purchase target (what's being bought) → <PURCHASE_TARGET>. Offer to seed context/spec.md with a short intake.
• general-research-workspace: the research question or theme → <RESEARCH_QUESTION>. This is the umbrella question; individual questions accrue under questions/.
• obsidian-vault: the central question → <RESEARCH_QUESTION>, plus the topic slug for the #research/<topic> tag.

Write these into CLAUDE.md or context/scope.md as appropriate.

6a. obsidian-vault only — substitute placeholders and seed the tag

This variant carries placeholders in more files than the others, because the vault's own notes are part of the scaffold. Replace across every .md and the .canvas:

Placeholder	Replace with
<WORKSPACE_NAME>	the workspace name
<RESEARCH_QUESTION>	the central question
<DATE>	today's date, ISO (YYYY-MM-DD)

cd <workspace>
grep -rl '<WORKSPACE_NAME>\|<RESEARCH_QUESTION>\|<DATE>' . --include='*.md' --include='*.canvas'

Substitute in each, then confirm none remain. Research Map.canvas is JSON — edit the string values only, and re-validate with python3 -m json.tool afterwards.

Then set the topic tag: replace the empty research/ tag in Meta/Tags.md, Home.md, Context/Scope.md and the six files in Templates/ with research/<topic>, so notes created from the templates carry it from the start.

Leave the {{date}} tokens in Templates/ alone — those are Obsidian's own Templates-plugin syntax, filled at note-creation time, not now.

7. Initialise git and (optionally) publish

cd <target-parent>/<workspace-name>
git init
git add .
git commit -m "Initial workspace from research-space plugin (<variant>)"

Unless --local-only is set:

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

8. Auto-register into Open-Research-Workspaces-Index

Skip this step if any of: --local-only, --private, --no-index.

The public index lives at github.com/danielrosehill/Open-Research-Workspaces-Index. Append a new entry to its README.md under the ## Research Workspaces section so the new workspace is discoverable.

Procedure:

1. Clone the index into a temp dir:

   TMP=$(mktemp -d)
   gh repo clone danielrosehill/Open-Research-Workspaces-Index "$TMP/index"
   

2. Edit $TMP/index/README.md. Insert the following block immediately before the --- that closes the ## Research Workspaces section (i.e. append as the last workspace entry, before the horizontal rule that precedes ## Templates):

   ### <Human-readable title derived from workspace name>
   
   <One-paragraph description>. **Question:** <the central question / umbrella theme>.
   
   [![View Repo](https://img.shields.io/badge/View-Repo-blue?style=flat&logo=github)](https://github.com/danielrosehill/<workspace-name>)
   

   Also bump the **Last Updated:** <Month YYYY> line to the current month/year if it's out of date.

3. Commit and push:

   git -C "$TMP/index" add README.md
   git -C "$TMP/index" commit -m "Add <workspace-name> to research workspaces index"
   git -C "$TMP/index" push
   

4. Clean up: rm -rf "$TMP".

If the insertion fails (section markers changed, merge conflict, etc.), report it and let the user handle manually — do not force-push.

9. Print next steps

Tell the user:

• Workspace path, variant, and GitHub URL.
• Whether the workspace was added to Open-Research-Workspaces-Index (and the link), or why it was skipped.
• First commands to run: /research-space:research-init, then /research-space:source-log <url> as they gather material.
• When they have a handful of sources: /research-space:summarize-sources.
• For focused investigation: /research-space:deep-dive <topic>.
• For general-research-workspace: note the Q&A pair convention (questions/<slug>.md + answers/<slug>.md) and that /research-space:export-report consolidates selected pairs into a single doc.
• For obsidian-vault: tell them to open the folder in Obsidian (Open folder as vault) — .obsidian/ is committed, so it opens configured — and to start at Home.md or Research Map.canvas. Point them at Meta/Conventions.md for the frontmatter schema. Mention that no community plugins are needed, and that Meta/Dataview Queries.md becomes live if they have Dataview.
• Reminder that the workspace is data — they can delete/move it freely without losing plugin commands.

Notes

• Resolve the scaffold via ${CLAUDE_SKILL_DIR}/../../template/ (not ${CLAUDE_PLUGIN_ROOT}).
• Never copy .claude/commands/, .claude/agents/, or .claude/skills/ into the new workspace.
• Don't hard-code personal paths or identifiers — everything comes from user memory or prompts.
• Prefer Train-Case repo names (State-Of-Claude-Context-0426) over kebab-case when publishing publicly — matches Daniel's convention.
• The index registration step is intentionally idempotent-friendly: if the entry already exists (detect by URL substring), skip the insert and just push a no-op log.