Autonomous ScreenMeet Verification Agent
Give the agent a test scope and return to verified results.
I can give Claude a prompt such as “test file transfer,” “run the full smoke test,” or “reproduce this issue on a specific client version” and walk away. The agent creates a real session, drives the CRM and viewer, operates a Windows endpoint, captures evidence, and cleans up when it finishes.
Why I built it
Verifying a ScreenMeet change used to mean manually coordinating several systems: create a CRM record, launch a session, connect the right Windows client, grant permissions, exercise the feature, inspect the result, save evidence, and tear everything down. Repeating that work for every release or reported defect was slow and inconsistent.
I encoded that product knowledge into a reusable Claude tool. Instead of prescribing every click, I state the outcome I need. The agent chooses the workflow, handles the cross-system steps, and returns evidence tied to explicit pass or failure signals.
What I can ask it to do
- Full release smoke test: run the complete 14-feature matrix across CRM, viewer, and endpoint surfaces.
- Focused feature check: test one capability, such as file transfer, remote control, screenshots, chat, or AI-generated summaries.
- Issue reproduction: recreate a reported failure against a specified ScreenMeet client version and collect the relevant evidence.
- Targeted depth: stop after a basic connection check or enable the required features and exercise the entire product path.
The same tool supports quick investigation and broad release confidence. I can start a meaningful test with one prompt while continuing other support or engineering work.
The verification path
- CreateCRM case or incident
- ConnectLive Windows endpoint
- ExerciseViewer features
- ConfirmEvidence and teardown
The workflow supports both Salesforce and ServiceNow while sharing the same viewer, endpoint, validation, and cleanup logic.
- Core session: streaming, remote-control consent, administrative state, and connection health.
- Support tools: clipboard, Windows Tools, System Info, chat, screenshots, file transfer, privacy curtain, and stream tuning.
- Applied AI: Ask and Agent modes, human approval before execution, execution results, and the generated session summary.
- CRM proof: screenshot upload, session records, status changes, termination reason, and summary submission.
State-aware testing
The tool models a session as a permission ladder. View-only, remote control, and administrator access expose different capabilities, so each assertion waits for the required grant and checks the state the endpoint reports.
It also distinguishes expected product variation from defects: a single-monitor endpoint should not show display switching, System Info can be empty in Live sessions, and AI Agent mode remains advisory until remote control is granted.
Evidence and cleanup
Each feature has an observable pass signal rather than a vague “looks good” step. Examples include a transferred file appearing on both sides, a Windows command producing a known marker, a screenshot landing on the CRM record, and the endpoint process disappearing after closure.
Teardown is part of the test: close the session through the correct path, confirm the viewer reports closure, verify the endpoint process ended, and remove stale viewer tabs so the next run starts from known state.
Skills used
Quality engineering: end-to-end test design, state machines, explicit assertions, regression coverage, and reproducible failure evidence.
Systems integration: Salesforce, ServiceNow, browser automation, WebRTC viewer behavior, Windows endpoints, AWS Systems Manager, and CRM record validation.
Agent design: turning product knowledge into bounded actions, permission gates, branching workflows, and reliable handoffs while keeping a human in control.