The Agent Skills Directory

Command-line interface fallbacks: The skill includes instructions to use Azure CLI (az rest) or Azure PowerShell (Invoke-AzRestMethod) if primary tools are unavailable. This involves executing shell commands which could be a concern if misused, but the skill provides specific guardrails to prevent the exposure of access tokens.
Handling of external pricing data: The skill generates short-lived URLs for pricesheet downloads. This is a potential external download concern, but it is mitigated by requiring the user to manually download the file rather than having the agent process the remote URL directly.
Indirect prompt injection surface: The skill ingests data such as Azure scopes and pricing files provided by the user (references/pricing-estimate.md). This presents a surface where untrusted data could influence agent behavior, although the risk is reduced by instructions to use server-side filtering and structured tool contracts instead of local parsing.
Credential safety instructions: The workflow includes explicit warnings against exposing access tokens or pricesheet URLs when performing API fallbacks. This addresses potential data exposure concerns by ensuring sensitive authentication information is not leaked into the conversation history.

cost-estimation