retro

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

• Third-party content exposure detected (high risk: 1.00). The skill's deep-mode workflow explicitly requires subagents to perform web searches and fetch external resources (Agent B "Learning Resources" uses the "Web Research Protocol" in {CWF_PLUGIN_DIR}/references/agent-patterns.md, retro-gates-checklist §Deep Mode Artifact Gates requires external web resources in Section 6, and references/expert-lens-guide.md mandates web search to verify experts), so the agent will ingest and act on open/public third-party content.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's deep mode explicitly requires web searches and fetching external web resources at runtime (Expert Lens: "Web search is REQUIRED" and Learning Resources: "find 2-3 external resources using the Web Research Protocol"), meaning arbitrary http(s) URLs discovered and retrieved during execution will be used to ground and directly control sub-agent prompts and outputs (i.e., "external web resources (arbitrary http(s) URLs discovered at runtime)").