superego

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). SKILL.md:35-39 downloads outsider-authored free text (AGENTS/prompt files) from raw.githubusercontent.com at runtime via curl, which can then be loaded into the skill’s LLM context.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The install/init steps download skill and agent prompt files from https://raw.githubusercontent.com/cloud-atlas-ai/superego/main/codex-skill/... (curl in SKILL.md and README.md), and those fetched markdown files directly control the agent prompts/instructions and are required for the skill to function.