cardano-cli-doctor

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • EXTERNAL_DOWNLOADS (MEDIUM): Metadata requests installation of system utilities (curl, docker, colima) via Homebrew which requires elevated privileges.
  • REMOTE_CODE_EXECUTION (MEDIUM): scripts/cardano-cli.sh pulls and executes the ghcr.io/intersectmbo/cardano-node image. While this is the official Cardano repository, the organization is not on the trusted whitelist.
  • COMMAND_EXECUTION (MEDIUM): Wrapper scripts enable the agent to run arbitrary cardano-cli commands using pass-through arguments ($@), providing a significant execution capability.
  • DATA_EXFILTRATION (LOW): The Docker wrapper mounts the current working directory ($PWD) and node socket directories into the container, allowing for potential file exposure or manipulation.
  • PROMPT_INJECTION (MEDIUM): The skill ingests un-sanitized output from command help text and version info into the agent context (SKILL.md context and scripts/cardano-cli-doctor.sh). Evidence: 1. Ingestion point: cardano-cli help output; 2. Boundary markers: Absent; 3. Capability inventory: Bash execution; 4. Sanitization: Absent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 01:07 AM