claude-code-guide

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

• Third-party content exposure detected (high risk: 0.85). Yes—SKILL.md’s required workflow includes runtime WebFetch of public Claude Code documentation pages (e.g., https://code.claude.com/docs/en/), which are outsider-authored free text that will be ingested into the agent’s LLM context for “Compare & Recommend.”

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly WebFetches documentation from https://code.claude.com/docs/en/ (e.g., https://code.claude.com/docs/en/memory, https://code.claude.com/docs/en/skills, etc.) at runtime and uses that fetched content to drive its recommendations and updates, meaning remote content directly controls the agent's prompts/behavior.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.70). The skill explicitly instructs the agent to "Edit existing config files" and "Create new config files" (Step 4) while also referencing system-managed locations like /Library/Application Support and managed-settings.json that may require elevated privileges, so it could modify system state or attempt privileged changes.