MOMM 1.9.0

MOMM 1.9.0 — guided setup and maintenance

Released 2026-08-23.

MOMM 1.9.0 turns first-run setup from a collection of terminal checks into a local, task-oriented Setup Center. The current coding agent remains the sole controller and writer; external CLIs remain untrusted, read-only reviewers.

What changed

Local Setup Center

• Added a browser-based Setup Center served only on 127.0.0.1.
• Added a separate Controller selector so the active driver is never presented as part of its own peer-review pool.
• Consolidated CLI installation/version, account-session evidence, model access, and the next action into one provider card for Claude Code, Antigravity, GitHub Copilot, and Grok.
• Added a real completion indicator with non-overlapping counts for installations, sign-ins, detected sessions awaiting verification, and provider CLI updates.
• Added Quick Setup, which refreshes local state and verifies detected sessions sequentially using harmless synthetic text. It never launches interactive OAuth automatically.
• Kept provider installation, sign-in, and updates as explicit, visible user actions.

Skills and diagnostics

• Added read-only checks for local versus published skill versions.
• Grouped skills into Update available, Modified locally, and Up to date.
• Added Review diff and Commit… terminal handoffs for local changes. These handoffs never stage or commit files automatically.
• Blocked automatic skill pulls whenever the repository has uncommitted changes; clean updates use git pull --ff-only.
• Added provider-native CLI update checks and model-access reporting.
• Reduced runtime noise to a one-line health summary and a collapsed diagnostics panel. Environment checks report relevant variable names only, never values.

Guided onboarding

• Added scripts/onboard.mjs as a zero-model-call terminal fallback for headless environments.
• Added a first-run walkthrough with exact setup states, recovery actions, rollout guidance, and source-sharing boundaries.
• Updated harness compatibility notes for the locally verified Claude Code, Antigravity, Copilot, and Grok CLI surfaces.

Trust boundaries

• Setup and maintenance never read credential contents, passwords, API-key values, or environment-variable values.
• MOMM remains OAuth/account-session only. Known API-key variables are stripped from reviewer processes.
• Setup sends no repository or project content. Optional connection verification sends only disclosed synthetic text.
• A real peer review does send the user-selected diff or input to the chosen provider. The user must confirm that source may be shared before dispatch.
• The controller remains the only writer and must reproduce material findings before applying changes.
• Reviewers cannot commit, edit the working tree, or authorize their own findings.

Deliberately not automated

• Completing browser OAuth or handling credentials.
• Opening multiple interactive sign-in flows from Quick Setup.
• Staging or committing local skill changes.
• Pulling over a dirty working tree.
• Installing or updating provider software without a visible user confirmation.

Verification

The release adds CI coverage for guided onboarding and the Setup Center safety contract. Before publication it passed:

• Setup Center syntax and deterministic self-tests;
• MOMM dispatcher deterministic self-tests;
• zero-call onboarding and input-safety checks;
• official Agent Skill structure validation;
• live loopback HTTP, maintenance API, and CSRF-protected action checks;
• the dirty-repository update guard; and
• git diff --check.

Start the graphical flow with:

node momm/scripts/setup-ui.mjs

For a terminal-only first run:

node momm/scripts/onboard.mjs --governor codex

← All version notes