skills/koyeb/koyeb-skills/databases/Gen Agent Trust Hub

databases

Warn

Audited by Gen Agent Trust Hub on Mar 26, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/koyeb-cli.sh acts as a pass-through wrapper (koyeb "$@") that executes the Koyeb CLI with any arguments provided by the agent. This grants the agent broad execution capabilities over the Koyeb environment.
  • [COMMAND_EXECUTION]: A complex shell command chain is documented in references/koyeb-cli.md which utilizes awk and xargs to perform bulk deletions (koyeb apps delete). This pattern poses a risk of unintended large-scale resource destruction if the filtering logic is bypassed or if resource names are manipulated.
  • [DATA_EXFILTRATION]: Documentation in references/koyeb-commands.md and references/koyeb-databases-flags.md promotes the use of the --debug-full flag. This flag is explicitly described as preventing the masking of sensitive information, such as API tokens, in the output, which could lead to credential exposure if the agent's output is logged or intercepted.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It ingests untrusted data from the user (such as database names and instance types) and interpolates them directly into shell commands without sanitization or boundary markers.
  • Ingestion points: User-provided database names, application names, and configuration flags in SKILL.md.
  • Boundary markers: None present; user input is directly concatenated into command strings.
  • Capability inventory: Full access to the Koyeb CLI via scripts/koyeb-cli.sh, including resource creation, modification, and deletion.
  • Sanitization: No validation or escaping is performed on external content before it is passed to the shell script.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 26, 2026, 04:50 PM