teach-oh

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly instructs the agent to fetch and install code from public GitHub raw URLs (e.g., the raw.githubusercontent.com links in Step 4, Step 5 and Step 6 for hooks and agent files), which are open/public third-party sources the agent will read and use during its workflow.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs at runtime to fetch and install remote code that controls agent behavior from raw GitHub URLs (e.g. https://raw.githubusercontent.com/open-horizon-labs/skills/master/hooks-omp/oh-skills-phase.ts and the agents URLs under https://raw.githubusercontent.com/open-horizon-labs/skills/master/agents-omp/ and https://raw.githubusercontent.com/open-horizon-labs/skills/master/agents-claude/), so external content would directly control prompts/agent behavior and be required for those optional install features.