cursor-laptop-skills

Slash commands for every Cursor Cloud Agent, on every repo, after one account install.

Laptop ~/.cursor/commands and ~/.claude/skills never copy onto Cloud VMs. Cloud / indexes:

1. Built-ins
2. This repo's .cursor/commands + .cursor/skills (this repo only)
3. Account plugins (all repos)

This plugin is (3). Do not vendor copies into product repos.

Slash commands

Folded-in skills (no separate plugins needed)

These came from three third-party packs, trimmed to the keepers. Do not install the packs for these workflows; if they are still installed, remove them so names do not collide.

Local edits: Codex/Gemini agent manifests dropped; /setup-matt-pocock-skills and "Skill tool" references replaced (tracker = GitHub via the GitHub MCP, else local .scratch/ files); handoff renamed handoff-brief because /handoff is already the 7-phase close above; i-have-adhd never hands you a command an agent can run itself.

Skills that back those slashes set disable-model-invocation: true on purpose — they load when you type /, not as silent always-on rules. Exceptions that are always-on:

• skills/engram-save/SKILL.md — mem_save during the session, not only at wrap-up
• skills/tailscale-cloud/SKILL.md — Tailscale userspace, JobHub on nicholass-imac, GitHub extraheader fix, Cloudflare tokens already on Cloud
• skills/hq-law/SKILL.md — the standing law from ~/AGENTS.md, which no container ever sees: done means done, unknown is never a value, every posted pick real-money safe, never tell Nick to run a terminal command

Model-invocable by description (also have a / command): tdd, diagnosing-bugs, codebase-design, domain-modeling, ponytail. Their descriptions are the only always-on cost.

Cloud HQ (every Cloud Agent, every repo)

Cursor Cloud does not run this plugin's mcp.json (Cloud enables the plugin as skills/commands only) and does not read a repo .cursor/mcp.json. MCP for Cloud Agents is account/team dashboard:

1. Paste the output of python3 scripts/cloud_hq_mcp.py --print-mcp (two servers: engram, tailscale-imac) as personal MCP (https://cursor.com/agents) and Team MCP (https://cursor.com/dashboard/integrations). This is the only recommended Cloud registration for Grok Bot and every other host too. Exact steps: docs/CLOUD_ACCOUNT_MCP.md. Re-paste after every release that changes scripts/: the JSON embeds them.
2. Disable the marketplace Engram plugin on Cloud Agents. It can conflict with the launcher: command: engram exits 127 on a VM that has not installed yet and leaves namespace Engram in error. The launcher is the Cloud Engram server.
3. Cursor User secrets TS_API_KEY / TAILSCALE_AUTHKEY when injected. If they are not, cloud_tailscale_hydrate.sh GETs private R2 prompt-betting-engram/tailscale.env with CLOUDFLARE_API_TOKEN. Do not ask Nicholas to re-paste a Tailscale key.
4. Product-repo "start": "bash scripts/cloud_agent_start.sh" is still useful (Engram on PATH + engram serve + Tailscale userspace) but it is per environment. The dashboard MCP is what follows a brand-new repo.

The launcher looks for the scripts in this order and never needs PLUGIN_ROOT: $CURSOR_LAPTOP_SKILLS_ROOT, ~/.local/share/cursor-laptop-skills, the newest ~/.cursor/plugins/cache/*cursor-laptop-skills* tree, ~/.local/lib/engram, ./scripts, /workspace/scripts. If none has them it writes the copies embedded in the pasted JSON to ~/.local/lib/engram and execs those. A time-bounded git clone (SPORT_REPO_TOKEN extraheader) is the last resort, never the first. python3 scripts/cloud_hq_mcp.py --plant (or scripts/cloud_plant_hq_mcp.sh) pre-seeds ~/.local/lib/engram and, for hosts that hard-code /workspace/scripts/cloud_tailscale_mcp.py, adds git-excluded symlinks there when the path is free. GitHub (SPORT_REPO_TOKEN) and Cloudflare (CLOUDFLARE_API_TOKEN) already inject. JobHub MCP is read-only; Tailscale MCP tools imac_exec / jobhub SSH to the iMac. eval "$(python3 scripts/cloud_github_git_env.py)" replaces Cloud's stale git bearer extraheader. iMac MagicDNS: nicholass-imac. Do not export HTTP_PROXY globally. Do not put keys in a repo.

Engram Cloud MCP (this plugin)

mcp.json launches scripts/cloud_engram_mcp.sh mcp --tools=agent with ENGRAM_CLOUD_AUTOSYNC=1. That is the same Gentleman-Programming agent profile (mem_save / mem_search / mem_session_summary). Do not add a second memory product (not engram-memory.com).

The wrapper self-heals: if ~/.local/libexec/engram is missing it runs cloud_install_engram.sh (Linux amd64/arm64 and Darwin amd64/arm64, sha256-pinned) then hydrates Cloud autosync. stdout stays MCP-clean. A missing binary must never leave the Engram namespace in error.

Cloud token and server URL are not in git. scripts/cloud_engram_hydrate.sh reads private R2 prompt-betting-engram/client.json with CLOUDFLARE_API_TOKEN (already injected on Cloud) and writes ~/.engram/cloud.json mode 0600. Worker: https://engram-cloud.nikhouseholdr.workers.dev (GET /health → {"status":"ok","service":"engram-cloud"}). The installer also writes global ~/.cursor/mcp.json from cloud_hq_mcp.py --print-mcp (desktop + Cloud) and, on macOS, a KeepAlive LaunchAgent for engram serve. Cloud Agents still need that same JSON saved in the MCP dropdown — ~/.cursor/mcp.json on the VM is not the dashboard.

A local SQLite without that hydrate is not the iMac store.

Tailscale iMac MCP (this plugin)

mcp.json also launches scripts/cloud_tailscale_mcp.py. Tools: tailscale_status, tailscale_up, imac_exec, jobhub. Initialize succeeds even when TS_API_KEY is missing — the tools then report the skip. It speaks MCP stdio and answers in the framing the host used: newline-delimited JSON (the MCP spec; Grok Bot) or Content-Length frames. Before 1.8.0 it always answered with Content-Length frames, which a newline-delimited host cannot read: initialize "timed out" while the script was healthy. The installer merges tailscale-imac into ~/.cursor/mcp.json next to Engram. Cloud hydrates TS_API_KEY from private R2 when the User secret is not injected. Never environment-scope those keys. Never put them in git.

Install (required once, on the Cursor account)

Until this is installed on the account, Cloud / will not list these commands.

1. Cursor desktop → Customize → Plugins
2. Add GitHub repo: https://github.com/nhouseholder/cursor-laptop-skills
3. Enable it for Cloud Agents (account or team), not only this workspace
4. Start a new Cloud session — the current one will not retroactively grow its / menu

Team Marketplace import of this private repo works the same if you share it with the team.

A sharp-oracle environment snapshot is not a substitute for Team MCP. Snapshots follow that environment; they do not follow diamondpredictions / mmalogic / other repos. The dashboard MCP does.

Update

Laptop remains canon when a skill file exists there. After a laptop edit: copy into skills/<name>/SKILL.md here, bump version in plugin.json and .cursor-plugin/plugin.json, push main. After any change under scripts/, run python3 scripts/cloud_hq_mcp.py --write-mcp .cursor/mcp.json (a test fails when it is stale) and re-paste the dashboard MCP. Cloud sessions pick up the new commit on next start.

Layout

commands/     # what Cloud `/` indexes (name + description frontmatter)
skills/       # the protocol each command reads (incl. folded-in i-have-adhd, ponytail, Matt Pocock set)
scripts/      # Tailscale, git extraheader, Engram install/hydrate, account-wide MCP launcher (+ --plant)
tests/        # python3 -m unittest discover -s tests
docs/         # CLOUD_ACCOUNT_MCP.md — dashboard paste for every Cloud session
mcp.json      # desktop/plugin Engram Cloud MCP + Tailscale iMac MCP (Cloud ignores this)
.cursor-plugin/plugin.json
plugin.json   # Agent Plugins manifest (skills + commands + mcp)