triage-finding
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- External Data Retrieval: The skill is designed to import findings from GitHub, Jira, and Linear. It uses REST APIs and specific connectors for this purpose. The instructions emphasize that all imported content—such as advisory descriptions or issue bodies—must be treated as untrusted data rather than instructions. This is a critical safeguard against indirect prompt injection.
- Secure Credential Management: Guidelines for GitHub intake follow security best practices. The skill instructs the agent on how to securely acquire tokens (e.g., via the GitHub Connector or standard environment variables) and explicitly prohibits printing or logging these tokens in the output.
- Indirect Prompt Injection Guardrails: The skill processes external, potentially attacker-authored text from vulnerability reports. To mitigate risks, the instructions include explicit 'Data Trust Boundary' sections and require the agent to ignore any commands embedded in the imported text.
- Evaluation and Calibration Harness: The repository contains a suite of Node.js scripts and configurations under the
evals/directory for testing and calibrating the skill's performance against benchmarks like SastBench. These tools utilize standard version control and package management commands, which is expected for development environments.
Audit Metadata