Mindatlas Auto Capture

Capture workflow:

1. Prefer mindatlas_submit_context_capture or another visible capture workflow when the user explicitly wants durable memory
2. When the chosen capability asks only for context, submit one high-value context block instead of assembling every final entry field yourself
3. In that context, include what happened, the result, why it matters later, and any clear time clues
4. Do not manually pass source, channel, session, or tool fields when the capability only asks for context; OpenClaw provides that request metadata automatically
5. Let MindAtlas materialize the final entry type, summary, content, tags, relations, merge, and dedupe behavior whenever the chosen capability supports that

Custom capability rule:

• Do not assume the only valid capture path is the built-in submit-context tool
• Administrators may expose custom capture workflows or agents
• If the visible dedicated capture tools are insufficient, route back through the dispatcher flow and inspect mindatlas_list_capabilities
• If the current session's MindAtlas capture surface looks stale after plugin, shipped-skill, or catalog changes, route back through overview/dispatcher and recommend npm --prefix ./integrations/openclaw-mindatlas run update:openclaw, then opening a new session

User control:

• If the user explicitly does not want something recorded, do not store it
• If privacy sensitivity is unclear, ask briefly instead of assuming
• If the information is only temporary or operational and does not belong in long-term knowledge, do not force it into MindAtlas