multiagent-protocol — setup wizard

1 Owner identity

The GitHub login of the person who will approve Quadrant D PRs and push break-glass commits. The bot trusts reactions/comments only from these logins.

Your GitHub login Additional allowlisted reviewers (one per line, optional) Display name (optional, used in bot diagnostic comments)

2 Supervised repositories

Which repos do you want the bot to gate? One per line, in owner/repo form.

Governance repo (where doctrine + Decision Inbox live) Supervised repos (one per line) Bot repo (leave blank to use governance repo)

3 Preferences — your working style

How your agents talk to you and how much they decide on their own. This is the personal layer — it goes into your private governance repo and is read by your agents, never by the bot. Everything here has a sensible default; skip fields you do not care about.

Language Language your agents speak to you Language for written deliverables (reports, docs) Same as primary Reports come back in the language above. English Deliverables always in English, even if you talk in another language. Bilingual Primary language and English side by side.

Communication Report style Conclusion first Verdict in the first sentence, details after. Detailed narrative A full walk-through is fine; no need to lead with the answer. How decisions reach you Options menu 2-4 labeled options with pros and cons. Good if you are not a developer. Direct question Plain questions are fine; you will weigh the trade-offs yourself. Batch questions Collect questions and ask them together instead of interrupting one at a time.

Autonomy

How much your agents do without asking. This does not change what the bot enforces — Quadrant D always waits for you — it only tunes how your agents behave below that line.

Cautious Confirm before anything outward-facing or hard to undo, even when the bot would allow it. Standard Follow the four-quadrant model as-is. Delegating Proceed on everything the bot allows and report afterward. Quiet hours (optional, 24h clock, e.g. 22:30-07:30) Timezone (optional, IANA name)

Taste ledger seed (optional)

Working rules you already know you repeat to agents — one per line. Each becomes a dated entry your agents follow from session one. You can keep adding to it later.

One rule per line

Vocabulary seed (optional)

Your shorthand — nicknames, acronyms, internal terms — so agents read your requests like a colleague would. One per line as term: meaning.

One "term: meaning" per line

4 Runner tier

Where does the bot's cron tick run? Most users start with the Free tier.

T1 — GitHub Actions Free Best for 1-3 supervised repos. No infrastructure needed. May exhaust monthly minutes around 4-5 repos. T2 — Self-hosted runner Required for 5+ repos. You provide a VPS, Raspberry Pi, or always-on machine. See docs/guide/self-hosted-runner.md. T3 — Paid GitHub Actions Pay-as-you-go Actions minutes. Simplest infrastructure but ~$10-50/mo for active fleet.

5 Built-in skills

Toggle the built-in skills. Leave defaults unless you have a specific reason to disable.

Hallucination guard Refuses to merge commits whose body references files that do not exist at the merged SHA. Recommended. Empty PR → Quadrant D A PR with zero file changes is treated as Quadrant D. Recommended (cheap defense against bot-bug or probe PRs). Break-glass auditor Required. Detects [break-glass-*] commits on main and demands ADR within 24h. Agent-* trailer validator (C5) Required. Every commit must declare which AI agent / model / session / machine authored it.

6 Agent registry

Which AI agents do you use? The bot accepts commits only from these agents' identities.

GitHub App slug (you'll register the App in Step 8; pick a lowercase hyphenated name)

7 Download config files

When you click Generate, the wizard creates the 6 config files. Either download as a ZIP and unzip into your repo, or copy each file individually and open the pre-filled GitHub "create file" link.

8 Register the GitHub App

After applying the config files, click below to register the GitHub App with the right permissions pre-filled. GitHub will open in a new tab.

This button activates after you generate config files (Step 7). The manifest is built from your inputs.