superego

Pass

Audited by Gen Agent Trust Hub on Jul 9, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The '$superego init' command in 'SKILL.md' fetches skill definitions and persona files from an external GitHub repository ('github.com/cloud-atlas-ai'). It also triggers the download and installation of the 'superego' binary through 'cargo' or 'homebrew'.
  • [COMMAND_EXECUTION]: The skill relies on executing the 'sg' binary for its core functionality. Additionally, the '$superego init' command performs various shell operations including directory creation, file downloads via 'curl', and local file manipulation of 'AGENTS.md'.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. The '$superego review' command reads content from local project files and git diffs to provide feedback. Malicious instructions embedded in the source code or commit messages of a project could potentially influence the AI's metacognitive evaluation.
  • Ingestion points: 'SKILL.md' (via the 'sg review-codex' command which processes file content and git history).
  • Boundary markers: None identified; the skill does not appear to wrap the processed data in protective delimiters or provide explicit 'ignore instructions' warnings to the model.
  • Capability inventory: The skill has the ability to execute the 'sg' binary and write instructions to the project's 'AGENTS.md' file.
  • Sanitization: No sanitization or filtering of external content was found in the provided implementation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 9, 2026, 08:07 AM
Security Audit — agent-trust-hub — superego