OverviewHistoryStatsSecurity
npx skills add ...
Documentation
SKILL.md
npx skills add alwaysmeticulous/skills --skill meticulous-cli-update
Check whether the Meticulous CLI (@alwaysmeticulous/cli) and skills are installed and up to date, and install/update them if not. Invoked at the start of every other Meticulous skill, since the CLI and skills are under active development with frequent changes and improvements.
npx skills add alwaysmeticulous/skills --skill meticulous-cli-update
The @alwaysmeticulous/cli package is under active development and ships frequent changes and improvements. Other Meticulous skills assume the meticulous command is on PATH and up to date, so run this skill once at the start of any Meticulous workflow.
This only needs to run once per conversation. If you've already run it earlier in this conversation, skip it — there's no need to re-check the version or re-update on every Meticulous skill invocation.
Using the MCP server instead of the CLI? Steps 1-4 are about installing/updating the meticulous CLI binary and its own local auth — neither applies when calling tools on the hosted Meticulous MCP server, which is already on the latest version and authenticates via the MCP connection itself, not meticulous auth login. Skip straight to Step 5: the installed skills (this document set) are a separate thing from the CLI/MCP tool itself and still need to stay current either way, regardless of which one you're calling tools through.
Full setup instructions — installing the CLI, connecting the MCP server, and installing these skills — live at app.meticulous.ai/docs/agents/setup.
This skill normally runs as a sub-step of another Meticulous skill. The install/update commands below (Steps 1, 3, and 5) are security-sensitive — they install packages and reach the network — so treat them as best-effort and non-blocking:
npm install --global @alwaysmeticulous/cli@latest" — and move on. A declined prompt is not a failure; it just means "do it later."meticulous command exists at all (Step 1) — if it's genuinely not installed and the user declines to install it, no Meticulous skill can proceed, so stop there.The read-only checks (meticulous --version, npm view …, meticulous auth whoami) are safe to run directly.
If the command is not found, the CLI is not installed. Install it globally (best-effort, see the note above):
If installing isn't whitelisted, recommend the user run that command themselves. Because no Meticulous skill can run without the CLI, this is the one case where you should stop and wait if the user declines — there's nothing to continue with. Once installed, re-run meticulous --version to confirm it's on PATH, and skip to Step 4.
If the installed version already matches the latest, skip to Step 4.
Otherwise, update according to how the CLI is installed (best-effort, see the note above — if updating isn't whitelisted, recommend the user run the appropriate command and continue regardless):
Globally installed (typical — which meticulous resolves to a path outside the current project):
Locally installed in the project (@alwaysmeticulous/cli appears in the project's package.json and which meticulous resolves inside node_modules/.bin):
If the update ran, re-run meticulous --version and confirm it matches the latest before proceeding.
Verify the user is authenticated with Meticulous and has a project selected:
Add --json if you'd rather branch on structured output — it prints authenticatedVia and selectedProject.
There are four outcomes:
Nothing to do — continue to Step 5.
Also seen as project API token (~/.meticulous/config.json), test-run API token, or credentials injected at request time (agent platforms that attach a bearer credential to outbound requests to app.meticulous.ai). These credentials are scoped to a single project, which is already pinned, so there is nothing to select — do not run meticulous auth set-project: with an API token it errors out (the token is bound to one project), and with injected credentials there is likewise nothing to choose. Continue to Step 5.
The command exits with:
Not logged in. Run
meticulous auth login, or set METICULOUS_API_TOKEN. In terminals without a browser, usemeticulous auth login --non-interactive.
Sign-in is browser SSO, so a human always has to complete it in a browser; which login command to use depends on where you're running:
On the user's own machine (a browser there can reach this machine's localhost):
This prints a login URL and then waits on a local callback server, so run it in the background, then surface the printed URL to the user and ask them to open it and complete sign-in. Once they do, the command finishes and stores the token, and you can continue.
Alternatively, ask the user to run meticulous auth login themselves — at their own terminal that opens the browser directly.
On a remote or sandboxed machine (cloud agent, SSH session, container) where a browser on another device can't reach this machine's localhost:
This uses the OAuth device flow: it prints a URL and a short code instead of waiting on a local callback. Run it in the background, then surface the URL and code to the user and ask them to open the URL on any device and enter the code. Once confirmed, the command finishes and stores the token.
Both forms skip the interactive project picker, so add --project "Organization/Project" when you already know which project to use — that logs in and pins the project in one step. Without --project, an account with access to several projects lands in case (d) below, so re-run meticulous auth whoami once login completes and handle whichever case it reports before continuing to Step 5.
In a CI-like environment with no human available at all, the alternative is an API token: set METICULOUS_API_TOKEN (or pass --apiToken) instead of logging in.
whoami succeeds and additionally logs (on stderr):
No default project set. Run
meticulous auth set-projectto choose one.
Project-scoped commands can't resolve a project in this state. It happens for accounts with access to several projects — including right after a --non-interactive or --device login, which skips the picker. List the options, then pin one:
Ask the user which one to use unless it's unambiguous (e.g. only one project is listed, or the repo clearly corresponds to one of them). Alternatively, ask the user to run meticulous auth set-project themselves — without --project it shows an interactive picker.
The selection is saved to the account rather than the machine, so it also applies to the MCP server and to the user's other machines. meticulous auth get-project prints the currently resolved project on its own.
The skills themselves are also under active development. How to update them depends on how they were installed (best-effort, see the note above):
Installed with npx skills — the default. Skill files live under .claude/skills/, .cursor/skills/, or the equivalent for the agent, alongside a skills-lock.json:
Only run this once you've seen a skills-lock.json (or the skill files themselves) in the project, since it would otherwise install a second copy alongside a plugin install.
Installed as the Claude Code plugin — the skills show up namespaced as /meticulous:<skill-name>, and claude plugin list lists meticulous@meticulous. npx skills won't touch these; update the plugin instead:
Tell the user the update only takes effect after they restart Claude Code (they can also do this from the /plugin menu themselves). One exception: if claude plugin list reports the plugin's scope as managed, it's pinned by their organization's managed settings — don't try to update it, just mention it to the user.
Installed from the Cursor plugin marketplace — there's no command to run; recommend the user update it from Cursor's Customize → Plugins.
If the applicable command isn't whitelisted, recommend the user run it themselves. Either way — whether it ran, or the user declined — proceed with the calling skill.