| name | mcpchecker-results-triage |
| description | Triage the results mcpchecker evaluation, classify the results and propose fixes for false positives.
|
mcpchecker results triage
Analyze the results file (refered in the doc as {mcp_results_json_file}).
If not specified in the request, use evals/mcpchecker/mcpchecker-obs-mcp-tools-out.json.
Steps
- run
{skill_dir}/scripts/failed-tasks-with-details.py {mcp_results_json_file} to list the failed results
- for each task, classify if it's true positive or false positive
- for false positives, assess whether the reason is caused by:
environment - missed assumption in the targeted system
judge - incorrect or insufficient judge definition
- PAUSE: present the user a table with the results and wait for confiramtion or clarification.
- Once results confirmed, propose changes in the task definitions to address failures marked as
judge.
Available scripts
scripts/failed-tasks-with-details.py - script to prepare the info about failed tasks enriched with data from the task definition.