orca-alert-triage
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external Orca Security API responses via MCP tools (e.g.,
get_alert). It uses clear boundary markers and structured output templates to present this data to the user. While it has the capability to write to the filesystem, its logic is designed to process security metadata rather than executing arbitrary instructions from the alert content. - Ingestion points: Orca Security MCP tools (
get_alert,get_asset_related_alerts_summary). - Boundary markers: The skill uses strict markdown templates and visual separators for different sections of the alert analysis.
- Capability inventory: Filesystem write access (via
Writetool) and MCP tool execution. - Sanitization: The skill structures the data into predefined fields (Verdict, Timeline, Blast Radius), reducing the risk of the model interpreting data as instructions.
- [DYNAMIC_EXECUTION]: The skill dynamically generates remediation code (Terraform, CloudFormation, Pulumi, or CLI scripts) based on the specific context of the alert. These scripts are written to the local filesystem for user review. The process is transparent, involves user interaction to select the format, and provides verification steps for the user to perform manually.
- [SAFE]: All external communications and tool calls are directed toward Orca Security's official infrastructure or well-known cloud provider consoles (AWS, Azure, GCP). No unauthorized data exfiltration or suspicious network patterns were detected.
Audit Metadata