skills/mukul975/anthropic-cybersecurity-skills/building-cloud-siem-with-sentinel/Gen Agent Trust Hub
building-cloud-siem-with-sentinel
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted telemetry from multiple cloud providers into a centralized SIEM, where automated playbooks can perform high-privilege actions based on that data.
- Ingestion points: Untrusted security logs enter the system through data connectors for AWS CloudTrail, Azure AD, and GCP, as configured in
SKILL.mdand processed by the KQL queries inscripts/agent.py. - Boundary markers: The KQL queries and Logic App workflows lack explicit boundary markers or "ignore instructions" delimiters when processing string fields from security logs.
- Capability inventory: The Logic App definition in
SKILL.md(Step 3) possesses the capability to modify Azure AD user states (accountEnabled: false) and update security incident comments based on entities extracted from logs. - Sanitization: The automation relies on standard incident entity mapping; there is no evidence of additional sanitization or validation of the contents within log fields (e.g., UserIdentityArn, RequestParameters) before they are used to trigger or inform automated actions.
Audit Metadata