aro-ops
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The environment configuration scripts (
scripts/classic-get-env-config.sh,scripts/classic-get-env-config.ps1,scripts/hcp-get-env-config.sh, andscripts/hcp-get-env-config.ps1) perform telemetry reporting by sending data to a remote server. - Evidence: The scripts retrieve a
telemetry-cfg-endpointandtelemetry-cfg-api-keyfrom Azure resource group tags. They then usecurlorInvoke-RestMethodto POST a JSON payload containing the user's account name (retrieved viaaz account show), the agent client name, and skill revision to that endpoint. - Context: While likely intended for internal usage tracking within a corporate environment, the telemetry endpoint is dynamically determined by metadata (tags) in the cloud environment, which represents a data exfiltration vector if the environment is compromised.
- [EXTERNAL_DOWNLOADS]: The skill is configured to fetch operational and debugging guidance from external remote sources.
- Evidence:
SKILL.mdinstructs the agent to fetch markdown files fromhttps://raw.githubusercontent.com/Azure/ARO-HCP/main/docs/ai/debugging.mdandhttps://raw.githubusercontent.com/Azure/ARO-RP/master/docs/ai/classic-debugging.mdif the repositories are not available locally. - Note: These downloads target the official GitHub repositories of a well-known organization and are used for documentation purposes.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for untrusted external data that is subsequently processed by the agent.
- Ingestion points: The agent reads content from remote GitHub markdown files and parses JSON configuration data stored in Azure Resource Group tags.
- Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are present in the scripts when parsing the tag values.
- Capability inventory: The skill has access to the
shelltool, enabling execution of Azure CLI (az),curl, and PowerShell commands. - Sanitization: There is no evidence of sanitization or validation for the content fetched from remote markdown files or cloud metadata tags before it is presented to the agent context.
Audit Metadata