The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from Kusto log queries and external documentation, which is a known vulnerability surface for agents.
Ingestion points: Raw log data returned by scripts/kquery.sh and scripts/kquery.ps1; external documentation fetched from GitHub as per instructions in SKILL.md.
Boundary markers: No specific boundary markers or 'ignore' instructions are provided for the Kusto query output.
Capability inventory: The skill possesses shell capabilities, allowing the execution of bash, zsh, and pwsh scripts, as well as network operations via curl and az CLI.
Sanitization: Log results are returned directly to the agent context without sanitization or filtering.
[EXTERNAL_DOWNLOADS]: The skill instructions specify fetching environment-specific debugging guides from remote URLs.
Evidence: Retreives content from https://raw.githubusercontent.com/Azure/ARO-HCP/main/docs/ai/kusto-debugging.md and https://raw.githubusercontent.com/Azure/ARO-RP/master/docs/ai/classic-log-search.md.
Note: These sources belong to the Azure organization, which is a recognized and trusted service provider.
[COMMAND_EXECUTION]: The skill implements its core functionality through a set of shell and PowerShell scripts that perform network requests and data processing.
Evidence: Utilizes scripts/kusto.*, scripts/kquery.*, and scripts/klinks.* to orchestrate az account get-access-token, curl, and Invoke-WebRequest operations.
Safety: The scripts use safe practices for parameter handling, such as jq --arg for JSON construction and printf '%s' in bash to prevent command injection.

aro-kusto