aquaria-cloudflare-ops

Warn

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill contains logic to update itself by downloading a shell script from a remote URL and piping it directly to bash. This pattern is found in commands/droid/aquaria-cloudflare-ops.md and commands/opencode/aquaria-cloudflare-ops.md, targeting https://raw.githubusercontent.com/Alcyone-Labs/agent-skills/main/install.sh.
  • [COMMAND_EXECUTION]: The skill is designed to execute various CLI tools including pnpm and wrangler for operations such as deployment, health checks, and managing Cloudflare Workflows. Evidence is throughout SKILL.md and the references/ directory.
  • [PROMPT_INJECTION]: The skill processes untrusted user input via the $ARGUMENTS and args variables to determine operational tasks and generate command lines. This represents an indirect prompt injection surface. Ingestion points are in the command definition files (commands/opencode/aquaria-cloudflare-ops.md, commands/droid/aquaria-cloudflare-ops.md, commands/gemini/aquaria-cloudflare-ops.toml). Capability inventory includes full shell command execution. No explicit sanitization or boundary markers for the user input are identified.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 1, 2026, 06:42 PM