How To

How-to

Task-focused recipes for running the Agent M harness — from a first install through tuning the memory layer and keeping the vault healthy. New here? Start with the tutorial, then come back for the specific task you're on.

Start here

1. Your first install — fresh clone to a healthy installed scratch project in ~5 minutes.
2. Install into a project — add the harness scaffold to an existing repo.

What's here

Install & configure

• Your first install — the end-to-end first run, two repos + a vault.
• Install into a project — add the scaffold to an existing repo.
• Configure a new project — the first-session detect → propose → approve → persist flow.
• Update an installed harness — pull a newer harness version into a project that already has one.
• Use per-project install — when to keep --scope project instead of migrating to user scope.
• Run without a vault — operate the harness with no MemoryVault configured.
• Choose a storage backend — (forthcoming, V5-1 part 5) select the memory engine's storage backend and confirm its plugin is installed.

Run the loop

• Cut a release — tag, changelog, and GitHub release via the ship-release skill.

Tune & maintain the memory

• Use auto-context in phases — recall budgets, save modes, and confidence thresholds per phase.
• Tune auto-orchestration — the toggles, thresholds, and cooldowns behind the SessionStart briefing and idle chain.
• Audit the vault — run the read-only vault lint and apply suggested fixes.
• Find missing note links — run the link-discovery audit and add the suggested [[wikilinks]].
• Use AgentMemory in any agent — read the vault (read-only) from Claude.ai · Gemini · ChatGPT · Antigravity.

See also

Reference · Architecture · Home