onboard

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill explicitly instructs the agent to pull and display verbatim writing samples and "actual excerpts" from connected emails/docs and to include validated user-provided text (copy-pastes) in generated files and chat output, which could force secrets to be handled and output verbatim.

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

• Third-party content exposure detected (high risk: 0.90). Yes—Phase 2’s “Scan All Connected Sources” pulls writing samples and business context from connected outsider-authored tools (e.g., Gmail/Slack/Notion/Drive messages and documents), and those excerpts are then kept “in context” and used to pre-fill Phase 3 and generate context/voice-dna.md/context/working-style.md, which the agent’s LLM reads and uses at runtime.