Sessions & Activity
Understand how connected agents use your knowledge. Sessions group recorded client activity, while activity entries show supported tool operations, results, and the files involved, giving you a practical way to trace a saved result or investigate a failure.
When an agent says it read a brief, saved a note, or proposed a change, this is where you can look for the recorded operation and then inspect the actual source or output.
Open the activity workspace
In the Mac app, open Agents → Sessions. Mission Control provides the recent overview; the session and activity views help you inspect individual work.
Use the refresh control after a new task. Confirm the same MindWiki account when the app shows a different history from the one you expect.
Sessions, connections, and pages mean different things
| Item | What it represents |
|---|---|
| Connection | An OAuth grant or API credential authorizing a client. |
| Session | A recorded MCP interaction context with client and activity information. |
| Activity entry | A recorded tool operation or supported service event. |
| Saved page | The actual knowledge file created or updated by the work. |
A session can end or expire while the client's grant remains authorized. Revoking a connection stops future access through that credential; it does not remove the pages already saved or erase content retrieved into another service.
Client labels and agent names help you find related work. The authenticated connection establishes authorization; a free-form agent name alone is not proof of a person's identity.
Read an activity entry
Depending on the operation, records can include the tool name, time, result or error, result summary, and files read or written. Use these details to understand the attempt and find the affected page.
An activity summary is an operational record. Open the actual document to evaluate its content. A successful write can save an inaccurate summary, and a successful search does not establish that the final answer used every candidate result.
Trace a saved result
- Find the time or session for the task.
- Locate the capture, write, or proposal operation.
- Read the recorded outcome and affected paths.
- Open the saved page or proposal.
- Check that it contains the expected result and source references.
For a research summary, compare the cited source pages with the conclusions. For a handoff, check completed work, evidence, unresolved questions, and next steps.
Investigate common failures
| Outcome | What to check |
|---|---|
| Authentication failure | Reconnect the client through its authorization flow. |
| Scope or permission failure | Inspect both the grant and vault agent permissions. |
| Page not found | Check account, path, title, and whether the page saved. |
| Edit conflict | Read the latest revision and reconcile it before saving. |
| Write timeout | Inspect the destination before retrying the same operation. |
| Proposal not ready | Inspect the affected pages and proposal readiness in the MindWiki review queue. |
A timeout means the client did not receive a result in time. It does not by itself prove that no write occurred. Reusing the same idempotency key with identical arguments helps the server recognize a retry of the same intended save.
Review activity with an agent
An authorized client can use mindwiki_get_agent_activity and mindwiki_list_agent_sessions, subject to activity permissions:
Review recent MindWiki agent activity. Summarize saved outputs and failed operations. Include the relevant page paths and distinguish attempted changes from completed ones.These tools return bounded recent results. They are useful for operational review, not a promise of an unlimited permanent forensic record or a transcript of every action in every application.
Keep records useful
Ask agents to use meaningful save messages and identify output paths. Save the conclusions and verification evidence in the page itself when you want durable project history. Activity tells you about the operation; a maintained handoff tells the next agent what the work means.
Do not put account sign-in keys, API secrets, or sensitive tokens into page content or tool arguments unnecessarily. Manage credentials through account controls and revoke a particular grant when it is no longer needed.
Next: Mission Control, Shared Agent Memory, and Vault Health & Permissions.