security-scan-agentlisted

# Security scan with AI and independent agent review Use this skill for the “adapters + AI triage + independent review” branch of the unified workflow: approved local scanning, host-AI triage of redacted findings, and a separate `security-triage` review. ## Workflow 1. Follow the unified workflow through bootstrap, toolchain validation, plan, preview, per-scanner consent, and host-AI sharing. Start one lifecycle in mode `scan-agent` and record every granted or denied consent. 2. Preview and run approved scanners, recording every preview, result, and skip in that lifecycle. Create the redacted payload and have the host model analyze it only after the separate AI-sharing approval. Record that exact assessment with kind `host_ai_triage` before delegation, including a detailed AI note for every finding at its zero-based `findingIndex`. 3. After host-AI triage is complete, ask separately: “May I give the bounded, redacted findings and AI triage to the dedicated `security-triage` agent for an independent review?” Do not delegate before an unambiguous yes. 4. Invoke `security-triage` with only the redacted scanner evidence and recorded host-AI triage. It may use primary advisory sources only when network access was approved; otherwise it must mark advisory claims as unverified. 5. Keep scanner evidence, host-AI assessment, and agent assessment as distinct sections. The agent may propose remediation but must not modify files. 6. Record the agent's