Best Practices

Use conservative defaults first, then widen Murph's scope after you have verified setup, channel access, and policy behavior.

Start narrow

Begin with selected messenger channels instead of all accessible channels. Expand the watched scope after a short test session.

Verify setup

Run a local health check after changing provider keys, Slack app settings, channel scope, identity, schedule, or policy.

murph doctor

Use conservative policy

Keep auto-send off until you have reviewed how Murph drafts, grounds, queues, and skips replies in your workspace.

Use yolo only after setup and early sessions behave the way you expect. It is intentionally permissive for action autonomy, but factual replies still need source grounding and relevant read-only tool use.

Review activity

Check Activity after each early session. Activity shows the run timeline, source-index recall, tool calls, draft creation, policy decision, and final outcome.

Treat Memory As Context

Murph's runtime memory is typed SQLite state scoped to the workspace, session, channel, and thread. Use it for continuity and thread context.

For latest, current, today, status, changed, or source-of-truth questions, rely on live connected sources. Murph should not answer fresh-state questions from stored memory alone.

Reconnect after scope changes

Reconnect Slack after adding app scopes or changing user-search consent. Slack search requires the search:read user scope, and reconnecting lets Murph store a fresh local user-search token.

Keep plugin work scoped

Create local extensions as scoped plugins first. Use source-edit runs only when a change must modify Murph core.