landing-report
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from remote pull request metadata and local configuration files.
- Ingestion points: Data entering the context includes PR branch names and metadata from the GitHub CLI (gh pr view), as well as local VERSION file contents.
- Boundary markers: The rendering instructions rely on the LLM to format internal variables into a dashboard without explicit delimiter-based isolation for untrusted text fields.
- Capability inventory: The skill identifies and uses the Bash tool for telemetry, state recovery, and version calculations through local framework scripts.
- Sanitization: The instructions mandate the use of jq to extract specific JSON properties. This structured extraction prevents raw, malicious payloads in metadata from being directly executed or interpreted as system-level instructions.
- [COMMAND_EXECUTION]: Several bash-based local binaries from the gstack suite (e.g., gstack-skill-start, gstack-next-version, gstack-skill-end) are executed. These operations align with the skill's documented purpose for session handling and versioning logic within the vendor's ecosystem.
- [DYNAMIC_EXECUTION]: The skill incorporates a mechanism to follow dynamic onboarding and consent instructions emitted by the start-up script. To mitigate spoofing, these instructions are only honored if they are direct results of the initialization command and match the current session ID.
Audit Metadata