secrets
Warn
Audited by Gen Agent Trust Hub on Mar 26, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill instructs the agent to create secrets by passing the sensitive value directly as a command-line argument using the
--valueflag inSKILL.md. This practice is insecure as it exposes credentials in cleartext to process listings, shell history, and system logs.- [COMMAND_EXECUTION]: Documentation inreferences/koyeb-cli.mdincludes a complex shell pipeline usingawkandxargsto perform mass deletions of applications (koyeb apps delete). This pattern facilitates high-impact destructive operations without individual user confirmation.- [COMMAND_EXECUTION]: The skill references and highlights the--debug-fullflag inreferences/koyeb-commands.mdandreferences/koyeb-secrets-flags.md. This flag explicitly disables the masking of sensitive information such as API tokens in command output, increasing the risk of accidental credential exposure.- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting potentially untrusted data from external sources and having destructive capabilities. (Evidence: Ingestion points includekoyeb secrets listandgetinSKILL.md; boundary markers are absent; capability inventory includessecrets deleteand theapps deletepipeline; sanitization of API output is absent).
Audit Metadata