azure-hosted-copilot-sdk

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

• Third-party content exposure detected (high risk: 0.80). The skill's required workflow (references/copilot-sdk.md and references/existing-project-integration.md) directs the agent to fetch and read public GitHub content (via context7-query-docs / context7-resolve-library-id and fallback to github-mcp-server-get_file_contents for repos like github/azure-samples), which are open third‑party sources the agent must interpret and can materially influence subsequent actions (selecting snippets, scaffolding/copying files), enabling indirect prompt injection.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs runtime fetching of SDK documentation and code via context7 tools and a fallback to github-mcp-server-get_file_contents (e.g., https://github.com/github/copilot-sdk and the azure-samples/copilot-sdk-service repo), and those fetched files are selected and injected into the agent’s context to drive prompts and examples, so external content can directly control agent instructions.