build-competitive-brief

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

• Third-party content exposure detected (high risk: 0.75). SKILL.md requires “public web research” as a workflow source category and describes extracting nuggets from those sources, which at runtime can ingest outsider-authored free text from fetched public pages into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill mandates loading and running the preflight script from ../user-context/SKILL.md at runtime ("$sales:user-context"), and the returned sales_preflight envelope is treated as authoritative for prompts, context mapping, and final obligations, so the fetched content directly controls agent behavior.