Skip to content

feat: add GCP Agent Security Audit skill (#160) - #203

Open
Mhmda1998 wants to merge 23 commits into
google:mainfrom
Mhmda1998:feature/gcp-agent-security-audi
Open

feat: add GCP Agent Security Audit skill (#160)#203
Mhmda1998 wants to merge 23 commits into
google:mainfrom
Mhmda1998:feature/gcp-agent-security-audi

Conversation

@Mhmda1998

Copy link
Copy Markdown

Closes #160

Description

Added the gcp-agent-security-audit skill including:

  • BigQuery diagnostic log auditing.
  • Modular SQL threat detection patterns.
  • Prompt injection, jailbreak, role override, and data exfiltration detection.
  • BigQuery ML anomaly detection support.
  • Google Cloud Pub/Sub alerting for high-risk findings.

cc @gaurika-analyst

Mhmda1998 added 23 commits June 21, 2026 00:43
…ort support

Added advanced diagnostic capabilities for agent reasoning paths and a new CLI command to export metrics to BigQuery for ML-driven evaluation
Query retrieves timestamps, agent IDs, and non-null retrieved documents, filtering based on specific patterns.
Updated the AgentSecurityAuditor to load threat patterns from SQL files in the 'patterns/' directory, with a fallback to default patterns. Enhanced the run_audit method to include the patterns used in the audit results.
Expanded the documentation for the GCP Agent Security Audit Skill, detailing its features, requirements, and usage instructions.
Refactor GCP Agent Security Audit Skill to enhance functionality and logging. Introduce new security patterns and improve log fetching and auditing processes.
Expanded the GCP Agent Security Audit Skill documentation with detailed features and examples for prompt injection, jailbreak, role override, indirect prompt injection, and data exfiltration detection. Updated requirements and usage instructions.
Updated SKILL.md to include new examples and detailed architecture flow for AI agent security monitoring.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Feature request: New skill "gcp-agent-security-audit" for proactive threat detection

1 participant