---
title: "21. Troubleshoot and use the quick reference"
slug: "ai-agent-troubleshoot-and-use-the-quick-reference"
updated: 2026-10-29T21:34:48Z
published: 2026-10-29T21:34:48Z
canonical: "docs.screenmeet.com/ai-agent-troubleshoot-and-use-the-quick-reference"
---

> ## Documentation Index
> Fetch the complete documentation index at: https://docs.screenmeet.com/llms.txt
> Use this file to discover all available pages before exploring further.

# 21. Troubleshoot and use the quick reference

## Common problems

| Symptom | Check first | Expected next step |
| --- | --- | --- |
| No AI Assist button | Feature entitlement, AI Assist setting, session creation time | Save the intended configuration and create a new session if required |
| Agent is greyed out | Client compatibility, connection, Remote Control, session AI Agent enablement | Read the tooltip; restore the missing prerequisite |
| Agent became Ask during custom-tool testing | Which connection has Remote Control | Finish/detach testing, restore viewer control, and check mode before sending |
| A new built-in tool is missing | Awaiting review, platform/filter, saved enablement | Review and save the decision; test in a fresh session |
| A custom tool is missing | Published version, enabled selection, archive/platform status | Publish the intended saved revision and confirm selection |
| The assistant asks to choose a target | Multiple plausible devices, processes, packages, or other candidates | Select a concrete target using its displayed context; do not approve ambiguity |
| A supposedly read-only operation asks for admin | Its execution role and the protected data it needs | Review the inline admin reason; privilege and risk are independent |
| Administrator access was granted but the plan still waits | Human-in-the-loop setting or warning acknowledgement | Review and approve the operation separately |
| Typed “yes” did not open UAC | The inline card's Request admin access action | Use the button; credentials and device approval use the native workflow |
| Typed “yes” did not approve an installer warning | The specific card's required warning checkbox | Review and acknowledge the measured warning on that card |
| Generated code did not appear | Organization allowance, current Code mode, pre-built coverage, safety refusal | Check the exact mode; an available pre-built tool is preferred |
| Switching to auto-run did not run an old script | Whether it was already held for review | Review the existing card explicitly |
| Software/driver tools disappeared | High-risk execution setting and tool opt-in | High must require review; enable only intended suite operations |
| KB returns no expected article | Provider permissions, source enablement/scope, publication/channel/language | Run Test search for a known article; distinguish empty from failed |
| Public search is slow or incomplete | Source scope, trigger, depth, provider outcome | Narrow the query or use a focused source; review partial/timeout notices |
| A draft “disappeared” after refresh | Whether a revision was actually saved | Browser authoring history is not unsaved-source recovery |
| Author with AI timed out | Interrupted message, retained draft, connection/provider | Retry explicitly, possibly with a smaller request; inspect any existing draft first |
| A test timed out | Whether the native job remains outstanding | Retrieve/inspect the earlier result; do not blindly repeat a changing operation |
| The device/server disconnected | Viewer connection and pending device prompts | Reconnect first; an old card or a visible transcript is not proof that a job did or did not run |
| No explainability file | Artifact flags, completed session/upload, storage/permissions | Follow [section 22](https://docs.screenmeet.com/docs/en/ai-agent-review-explainability-and-export-evidence); absent records cannot be recreated from a flag enabled later |

When reporting a problem, include the session reference, time, selected mode, relevant saved settings, exact visible error, and whether an operation actually dispatched. Use the appropriate customer or internal explainability artifact. Avoid putting credentials, private screenshots, or unredacted internal logs into a broadly shared report.

## Quick reference

| Control | Purpose |
| --- | --- |
| AI Assist / Open AI Assist | Open the session's AI panel |
| Ask | Discuss and explain; an ordinary message does not execute device tools |
| Agent | Carry out eligible tool-based investigation/work |
| + / Add screenshot or image file | Attach a supported image or capture the remote screen |
| Code menu | Choose session script generation behavior within organization allowances |
| Stop generation | Stop further response/turn processing where cancellation is possible |
| Start over | Reset conversation working context; does not undo device changes or reset Code mode |
| Execute Plan | Approve selected steps of the current ordinary plan |
| Why this tool? | Show the selection rationale |
| View execution details | Inspect operation source, arguments, and available outcome details |
| Chat about this | Attach selected source/output to a question |
| Request admin access | Open the existing device privilege workflow |
| Continue without admin | Record the decision and continue with available alternatives |
| Review script / View script | Inspect generated code before/after its decision |
| Run script / Don't run | Resolve a script's review decision |
| Show tools awaiting review | Find new managed capabilities that remain off pending a decision |
| Keep disabled | Record that the administrator reviewed a tool and intentionally left it off; save the selection |
| Save revision | Persist an immutable custom-tool revision |
| Publish and enable | Roll out the latest saved custom-tool revision |
| Device testing → Run test | Execute the current tool-editor buffer on a connected real device |
| Send result to assistant | Stage exact test evidence for the next authoring message |
| Save skill | Persist the procedure draft; new organization skills still start off |
| KB Test search | Perform a real knowledge-provider query without an AI support answer |
| Web Run test | Perform a real scoped public search with the current source settings |