skill-installer

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

• Third-party content exposure detected (medium risk: 0.30). scripts/list-skills.py fetches and parses a GitHub directory listing (names) from https://api.github.com/.../contents for the user-supplied --repo/--path at runtime, which ingests outsider-authored free text from arbitrary repos.

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

• Potentially malicious external URL detected (high risk: 0.90). The install/list scripts fetch and install skill code from GitHub at runtime (e.g., they call https://api.github.com/repos/{repo}/contents/{path}?ref={ref} and download zips from https://codeload.github.com/{owner}/{repo}/zip/{ref}), pulling remote code that will become agent-executable and can control prompts/behavior.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly tells the agent to "request escalation" when running its networked install scripts in the sandbox, i.e., to seek elevated privileges, which encourages compromising the host environment.