site-builder

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill asks the agent to accept user-provided secrets and write them verbatim into project config files (e.g., paste a GitHub PAT into .claude/settings.local.json, place API keys into .mcp.json, and have analytics-agent "inject" credentials), which requires the LLM to handle and output raw secret values directly.

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

• Third-party content exposure detected (medium risk: 0.30). During Phase 1 DISCOVER the orchestrator spawns discovery-agent and injects “any user-provided context (URLs, documents, instructions)” while that agent “reads the repo” and “analyzes competitors,” meaning outsider-authored free text can be ingested to drive downstream doc creation and edits.