build-health

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

• Third-party content exposure detected (high risk: 0.85). Yes—scripts/fetch-builds.sh downloads outsider-authored Azure DevOps build timelines and failure log text from dev.azure.com into build-data/timelines/*.json and build-data/logs/*.txt, and scripts/analyze-builds.mjs reads those files and embeds extracted error excerpts into the generated markdown report that the agent then summarizes in chat.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly tells the agent to "request unsandboxed execution if sandboxing is enabled," which is a directive to bypass sandbox/security protections; otherwise the actions are limited to running fetch/analyze scripts in /tmp and using az/node (no sudo or system-file edits), so it is a high but not absolute risk.