kongctl-declarative
Warn
Audited by Gen Agent Trust Hub on May 15, 2026
Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The
SKILL.mdfile contains instructions for the agent to "request an unsandboxed retry" if a command is blocked by the execution environment. This is an explicit instruction to encourage bypassing security constraints or sandbox boundaries. - [COMMAND_EXECUTION]: The skill makes extensive use of the
kongctlCLI to manage cloud resources. This includes potentially destructive operations such assync(which deletes unmanaged resources) anddelete(which removes specified resources). - [EXTERNAL_DOWNLOADS]: The
references/cicd-github-actions.mdfile provides GitHub Actions examples that download and install the vendor's own tooling from GitHub (e.g.,kong/setup-kongctl@v1andkong/setup-deck@v1). - [DATA_EXFILTRATION]: The troubleshooting guide in
references/troubleshooting.mdinstructs the agent to inspect environment variables usingenv | grep '^KONGCTL_'. While targeted at vendor-specific configuration, this facilitates the exposure of local environment state and potential tokens.
Audit Metadata