implementing-msh

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md Step 3/4 can fetch upstream README/MIGRATION.md from GitHub raw URLs at runtime (outsider-authored public web content) and then use that fetched text to adapt/summarize and write code, which would be ingested into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill explicitly instructs the agent to fetch upstream docs at runtime (e.g. https://raw.githubusercontent.com/beecode-rs/msh-/main/README.md) via WebFetch to adapt behavior and write code for the user's installed major version, so remote content is fetched at runtime and directly controls the agent's instructions.