memforge-setup — Claude Code skill: MemForge routing
memforge-setup is a Claude Code skill for configuring MemForge workspace routing. It runs on requests to bind or remove a project binding, or to diagnose workspace selection.
Apache-2.0 ★ 104 updated 2026-11-17
/plugin marketplace add dodo-labs-ai/mem-forge
/plugin install memory@memforgeWhat memforge-setup does
The skill inspects and changes MemForge workspace routing for Git repositories and ordinary directories. Git bindings use a normalized origin; directory bindings use absolute paths, with the most specific directory taking precedence. An explicit workspace_id on an MCP tool call overrides either binding.
Python 3.12+ and access to the configured MemForge workspace directory are required. The scripts/workspace_config.py helper previews changes first; applying them requires a separate confirmation. Its commands accept codex or claude-code. The skill leaves MCP registration and credentials untouched.
What it helps with
- bind a Git repository to a MemForge workspace
- assign a workspace to an ordinary directory
- inspect which workspace a project resolves to
- remove obsolete binding keys
Example prompts for the agent
- Bind this Git repository to the selected MemForge workspace.
- Check which MemForge workspace the current directory resolves to.
- Show the removal plan for obsolete binding keys and wait for my confirmation.
How to install memforge-setup
1Plugin marketplace
/plugin marketplace add dodo-labs-ai/mem-forge/plugin install memory@memforgeThis installs the whole memory plugin with all its skills
2Skills CLI (Vercel)
npx skills add dodo-labs-ai/mem-forge --skill memforge-setup -a claude-code3Manually: put the files in place
~/.claude/skills/memforge-setup/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/memforge-setup/
1Skills CLI (Vercel)
npx skills add dodo-labs-ai/mem-forge --skill memforge-setup -a codex2Manually: put the files in place
.agents/skills/memforge-setup/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add dodo-labs-ai/mem-forge --skill memforge-setup -a cursor2Manually: put the files in place
.cursor/skills/memforge-setup/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install dodo-labs-ai/mem-forge integrations/claude-code/memforge-memory/skills/memforge-setup2Manually: put the files in place
.github/skills/memforge-setup/SKILL.md1Skills CLI (Vercel)
npx skills add dodo-labs-ai/mem-forge --skill memforge-setup -a gemini-cli2Platform CLI
gemini skills install https://github.com/dodo-labs-ai/mem-forge --path integrations/claude-code/memforge-memory/skills/memforge-setupManually: put the files in place
.devin/skills/memforge-setup/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about memforge-setup
How do I install memforge-setup in Claude Code?
Add the marketplace with /plugin marketplace add dodo-labs-ai/mem-forge, then install it with /plugin install memory@memforge.
What does memforge-setup do?
It inspects and configures MemForge workspace selection for Git repositories and ordinary directories. Changes are previewed before they are applied.
What does memforge-setup require?
It requires Python 3.12+ and access to the configured MemForge workspace directory.
Source text
SKILL.md · Apache-2.0
--- name: memforge-setup description: Inspect and configure MemForge workspace routing for Git repositories and ordinary directories. Use when the user asks to bind a local project to a workspace, change or remove that binding, or diagnose workspace selection. compatibility: Requires Python 3.12+ and access to the configured MemForge workspace directory. --- # MemForge Workspace Setup Use the bundled `scripts/workspace_config.py` helper. It owns only `~/.memforge/workspace-bindings.json`; never ask the user to edit the file by hand. ## Routing model - A Git checkout with an `origin` is stored in `repository_bindings` using the normalized origin, so its worktrees and moved checkouts share a binding. - Any ordinary non-Git directory is stored in `directory_bindings` by absolute path. Descendants inherit it and the most-specific directory wins. - An explicit tool `workspace_id` overrides either binding for that call. - Bindings select the workspace for the MemForge MCP tools. Hooks select no workspace: the plugin's only hook adds static usage guidance. - API URL and credentials remain in the client's existing configuration. This skill never writes or displays credentials and never edits MCP registration. ## Commands Resolve the helper relative to this skill, then use: ```bash python3 "<skill-dir>/scripts/workspace_config.py" --client CLIENT inspect --path "$PWD" python3 "<skill-dir>/scripts/workspace_config.py" --client CLIENT bind --path "$PWD" --workspace WORKSPACE python3 "<skill-dir>/scripts/workspace_config.py" --client CLIENT unbind --path "$PWD" python3 "<skill-dir>/scripts/workspace_config.py" --client CLIENT remove-obsolete-keys ``` `CLIENT` is `codex` or `claude-code`. Binding commands validate the workspace against the live configured target. Mutations are previews unless `--apply` is present. ## Obsolete keys `inspect` lists `obsolete_keys`: entries that earlier plugin releases wrote and nothing reads, such as `hook_workspace_id`. They are ignored and do not change routing. When the list is not empty, tell the user and offer to remove them with `remove-obsolete-keys`, following the confirmation workflow below. Never remove them without that confirmation. ## Required confirmation workflow 1. Use `list_workspaces` when available and let the user choose the workspace; never infer it from the server default. 2. Run the selected command without `--apply`. 3. Show the operation, binding kind/key, workspace, target origin, exact file, and planned document. Do not show credentials. 4. Ask for explicit confirmation of that preview. 5. After confirmation, repeat with `--apply --expect-digest DIGEST`, using the exact `document_digest` from the preview. 6. If the digest changed, preview and confirm again. Do not bypass the guard. An initial request to set up routing is not confirmation of the generated preview. Inspection is read-only and needs no confirmation. ## Completion report Report the local path or repository identity, workspace, binding source, file changed, and that the next MCP call will use it. State that no credentials or MCP registration changed. A restart is not required.
© MemForge Team · Apache-2.0 · dodo-labs-ai/mem-forge