Run inventory
Shell lists installed programs and saves the output.
You run a command on your PC to list installed software. The agent reads the output and matches it against public CVE records. You decide to patch critical wifi router vulnerabilities. It writes a CSV file with scores and priorities. You review the diff and accept the changes.
Shell lists installed programs and saves the output.
read_file loads the inventory and web_search finds CVEs.
write_file creates the CSV with scores and actions.
Ask first shows the diff for your review before saving.
list installed softwaresavedCVE for sample-app1 matchpatch-triage.csvcreatedpatch-triage.csvdiff shown# Sample triage data
id,software,version,cvss,priority,action
CVE-XXXX-0001,sample-app,1.0,7.5,High,update
CVE-XXXX-0002,web-server,2.1,4.3,Medium,monitor
CVE-XXXX-0003,db-core,3.0,9.1,Critical,patch
CVE-XXXX-0004,api-gw,1.5,2.0,Low,defer
CVE-XXXX-0005,auth-lib,0.9,6.5,Medium,review
CVE-XXXX-0006,net-lib,1.2,5.0,Medium,updateThis CSV file lists placeholder identifiers and invented software names. It includes a header row and six data rows with scores and actions. Scores are base scores, not your risk.
The agent runs your inventory command and saves the output to a text file, so it uses web_search to find relevant records for each product. Read the file. The read_file tool loads your inventory data for detailed analysis. Grep searches for matches in the records with high accuracy. You see the command results and the text files in the chat. Help understanding a cve entry comes from reviewing these specific details. The agent writes the findings to a new CSV file using write_file. You review the file in Ask first mode to check cvss cyber security matches yourself.
The agent edits the triage file to record correct scores and so you can verify the changes with a diff before saving the final version. Check the exit code. The agent uses edit_file to update the triage file with the correct CVSS scores. It shows a diff of the changes so you can verify the data. You can use Plan only to see the steps before they run. The agent saves the final file using write_file to complete the task. Common vulnerabilities and exposures are listed in the record for review. PHP object injection vulnerability explained helps you understand the specific risks.