# Activity and findings

Follow an investigation performed by your own agent.


![Platform demonstration: Follow the investigation in Activity.](/docs/screenshots/activity-light.webp)

## What opens a problem

Runtime classification and monitor rules decide when evidence becomes an incident. Repeated occurrences are grouped; the incident tracks counts, state changes, affected contexts, and linked evidence.

## Activity

The timeline records detections, comments, findings, code-change context, and agent progress. It is the visible record of an investigation, not a separate internal AI chat subscription.

A connected agent calls `get_incident_context`, publishes notes through `add_incident_note`, and updates phases through `update_incident_progress`. An active shimmer means the reported phase is active; agents should end that phase on completion or failure.

## Findings and evidence

Findings explain the investigation. Evidence can include stack traces, request failures, related releases, affected sessions/replays, device/browser surfaces, and impacted-user trends. Coverage depends on capture and permissions; missing evidence is not fabricated.

![Platform demonstration: Browse the evidence published during the investigation.](/docs/screenshots/findings-light.webp)

## Repair and verification

Your agent works in its own environment with its existing coding tools. Proposals, approvals, hosting operations, and GitHub actions are separate capabilities. An incident marked resolved is not a guarantee that an arbitrary remote deployment has already been fixed.

Configure [notifications and handoff](/docs/platform/notifications) if you need an external signal when new problems occur. A local agent that is not running cannot receive a live tool call.