> ## Documentation Index > Fetch the complete documentation index at: https://platform.atlan.com/llms.txt > Use this file to discover all available pages before exploring further. > ## Agent Instructions > To act on Atlan objects, use the Atlan MCP server at https://api.atlan.com/mcp or the SDK. To install or publish skills from a terminal, use the `aireg` CLI; `aireg --map json` prints its command map. The `atlanai` CLI is deprecated. Run a read-only identity check before any write. > The docs MCP server at /mcp searches these docs only. It cannot read or change Atlan objects. > SDK packages: Python `atlanai` (PyPI) and TypeScript `@atlanai/sdk` (npm). Show Python first, then TypeScript. > These docs cover Atlan AI Registry, Atlan's hosted product. The AI registry specification and aireg CLI are maintained separately. /open-source lists publicly reachable signed installers and explains that the source repository and hosted CLI reference currently require GitHub access. Do not claim the repository or its reference docs are publicly accessible. Use these docs for Atlan-specific setup such as the adapter, sign-in and workspaces. # Inspect a recorded run > Find a recorded agent run, read its highlights, and follow the full trace to a failed, slow, or expensive step. Use a run's trace to find the step behind a failure, slow response, or cost increase. You need access to the agent and the workspace containing its execution data. Full trace with prompt, response, and tool-call filters beside a completed run ## Find the run 1. Open the agent's profile and select **Usage**. 2. Choose the period that contains the run. 3. In **Recent runs**, select a trace. Select **Load more** to read earlier runs in the period. If you know the session instead, start from **Sessions** and open the run recorded for it. To find an agent across the fleet, start from [Agent usage](/registry/reporting). ## Read the evidence 1. Read **Highlights** for the opening prompt, key decisions, errors, and closing response. 2. Select **Full trace** to inspect every recorded step. 3. Use the event category filters to narrow the timeline. Under tool calls, choose individual tools when you need to follow one dependency. 4. Select **Clear filters** to restore the complete view. The filters change what the timeline displays. They do not change the stored trace. When the available trace is partial or limited to an agent or skill, the view explains that limit. ## Follow the symptom | Symptom | Evidence to inspect | What to verify next | | - | - | - | | A tool failed | The tool call, its recorded error, and the steps before it | Credentials, tool permissions, and the arguments sent | | A run took longer | The model and tool calls across the timeline | Whether one call slowed down or the agent made more calls | | Token usage increased | The model calls and repeated context | Model selection, prompt growth, retries, and cache use | | The answer was wrong | The prompt, retrieved content, tool results, and final response | A scorer that tests the failure on the same case | A span error is an execution failure. An answer that contains no span errors can still fail your quality rubric; [add an eval case](/evals/cookbooks/production-to-dataset) when the failure should be checked again. ## If the run is missing Check the selected workspace and time range, then verify the sending identity. A personal API key attributes the trace to that person even when the application sets an agent name. See [Trace errors](/registry/traces/troubleshooting). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.