INSTALL OR UPGRADE

One line.
Your agent does the rest.

Paste this into Claude Code, Codex, Antigravity, Copilot or Grok. It works for a new installation and for an upgrade.

Paste into your coding agent
Install MOMM for me by following https://marroccofella.github.io/skills/momm/install

What happens next

  1. It looks first. Your agent finds any existing MOMM, your harness and your version, and tells you what would change.
  2. It verifies before it runs anything. The release is checked against its signed tag and package hash. If the verifier (gitsign) is missing, it stops and asks you; it never skips the check.
  3. It shows you a dry run and asks. Nothing is installed or replaced until you say yes.
  4. It keeps what is yours. Reviewer logins, other skills and your private review ledgers stay as they are. It never asks for an API key.

Prefer to read before you paste? The complete installation guide · How updates and rollback work · Every version

The procedure your agent follows

This is the full text behind the line above. An agent that opens this page should follow it exactly and completely.

Two things agents ask. Which harness? The one you are running in now; if you cannot tell, ask the person. Is the verifier present? Run gitsign --version. If it is missing, that is normal on a first install: say in one or two plain sentences what gitsign is and that it is a one-time step, ask once, then install it exactly as Getting the verifier says for this system, Windows included. Never continue without it and never fetch it from anywhere but the official sigstore/gitsign release.

Install and upgrade procedure
Install MOMM if absent, or safely upgrade my existing installation to the latest publicly released
stable version. Do not install a candidate, an unsigned development head or a
version merely advertised by a page.

First detect my operating system and locate any actual installed MOMM skill (including the older
multi-llm-review name), its permanent repository clone, my current harness and
version. If absent, choose a permanent clone location outside my projects. Check the canonical marroccofella/skills release, manifest and release
notes, and show what changed since my installed version. Do not assume this
project directory is the skills clone or guess my harness installation targets.

Preserve my project, local changes, private review ledgers, reviewer logins and
existing skill scopes. Never request API keys, discard changes, overwrite an
unrelated discovery path or install tools in the background.

Before executing any newly fetched release code, verify its signed tag, expected
release-workflow signing identity and package hash. This applies to both an
ordinary update and a legacy bootstrap. Stop if verification is unavailable;
ask before installing a missing verifier.

Read the maintained new/legacy bootstrap guide at
https://github.com/marroccofella/skills/blob/momm-1.17.3/momm/references/bootstrap.md.
Do not mistake GitHub's bad_cert/Unverified badge for a gitsign verification result.
If gitsign is missing, no signature check ran. That is normal on a first install:
explain in one or two plain sentences what gitsign is and that it is a one-time step,
ask once, then install it exactly as the guide's "Getting the verifier" section says for
my system (Windows included), from the official sigstore/gitsign project only, by the route the bootstrap guide lists for my system. If a real signature, identity, transparency or
hash check fails, stop. A verified commit badge does not replace these checks.
Older releases lack bootstrap.mjs: obtain and independently inspect/trust the
small standalone helper before executing it, or use an organization-reviewed
copy. Never execute an unverified candidate's code to verify itself. Use its
offline --check first; explicit --prepare verifies an exact published stable
release into a new permanent directory without installing or replacing anything.

If my installation has the explicit update command and a valid momm.lock,
run the information check and then the verified dry-run from the skills clone.
Show the exact target version, changed files, complete protocol/default/persona
diff, saved harness scopes, network destinations and rollback path. Verify the
signed tag, expected signing identity and package hash. Ask me before applying;
ask separately for protocol acceptance if it changed. Never use --yes to bypass
protocol consent or fall back to git pull when verification fails.

Exception requiring an approved bootstrap, not a verification bypass: 1.15.0's
updater cannot open its bare signature-staging repository. Obtain a separate
normal clone at a published signed 1.15.1-or-newer tag; independently verify its
signer and package hash before executing its updater. Use that verified updater's
`--repo <existing-skills-clone>` option for the preview and explicitly approved
apply, preserving the existing receipt and exact scopes. Do not change the old
tag or claim that ordinary 1.15.0 self-update works.

For a new installation, read the verified release's installer help. Install only
MOMM into this harness's documented user-level skill scope: preview with
`node momm/scripts/install.mjs --target <chosen-harness> --dry-run`, then ask
before applying the same command without `--dry-run`. Do not use the repository-wide
installer's default to install unrelated skills. For an unnamed harness, verify
its documented custom skill directory; never invent paths or claim universal support.

If my old installation lacks the updater or receipt, explain that it needs a
one-time bootstrap. Show the exact permanent clone/ref/link changes and preserved
old installation before asking me to approve them. Do not invent a receipt,
silently retarget links, trust an old unsigned tag retroactively or claim rollback
is available before verifying it. Ask before installing a missing verifier.

For an existing momm discovery entry, prefer the separately trusted
migrate-legacy.mjs helper and its bootstrap.mjs dependency if available: preview
the exact prepared clone, entry, outside-discovery backup and complete protocol;
only after my separate approvals use --apply with the preview's --plan-sha256
and, if needed, --accept-protocol.
Save its journal and rollback command. Never put backups inside skill discovery,
or copy/overwrite project .ensemble_reviews ledgers and .gitignore into a new
clone. Verify a removed privacy rule by inspecting actual enforcement, not by
counting sanitization keywords. Unsupported aliases need inspection, not guessing.

After an approved install or upgrade, verify the installed version and harness discovery.
Read the newly installed MOMM skill, relaunch only the MOMM Setup Center process
you started, and show its local URL. Explain how to check reviewer versions and
explicit update actions. Distinguish login presence from a successful review;
ask before sending a synthetic connectivity test or any project source.

Make MOMM my default peer-review workflow across projects using this harness's
documented user-level standing instructions. Before writing it, show me the exact
section and where it goes, and wait for my yes. Preserve every existing rule and add
a labelled section: use MOMM for requested peer reviews and before finalizing
substantive coding changes, fixes, refactors and releases. The current agent stays
governor and sole writer; reproduce material findings, verify fixes and adjudicate
every suggestion. Respect project restrictions and provider-sharing permissions;
never recursively dispatch or run reviews for every conversational message.
Discovery links alone do not activate this rule. Verify it in a fresh context
where possible, or explain the necessary restart and any unsupported global scope.

From my chosen project, demonstrate a small approved review using the absolute
installed dispatcher path. Reproduce findings, decide every suggestion, verify
completion using the installed governor command, and give me the private ledger
link. Finish with the before/after version, the verified public release-note link,
test results, any unavailable routes and the verified recovery instructions.