See Also
- Agent Harness Landscape: dated product map, evidence states, and selection protocol
- Architecture: Claude Code's master loop, tools, context, agents, permissions, and MCP integration
- Tools Reference and Hooks Events Reference: exact runtime surfaces
- Context Engineering and Memory Systems: prompt assembly, persistence, retrieval, and drift
- Agent Evaluation and Observability: acceptance evidence, traces, metrics, and failure analysis
- Security Hardening and Native Sandbox: prompt-injection defense and execution boundaries
- Cross-Session Messaging: commit-bound handoffs, correlated-drift controls, and shared-working-tree risks
- Agent Teams and Agentic Software Factories: multi-agent coordination above one loop
- Repository Harness Engineering: project-level instructions, setup, state, and verification gates
- Machine-Readable References: release history, topic anchors, and normalized harness data
Last updated: August 2026. Claude Code implementation mapping checked through v2.1.250. Practitioner quotations checked against the yt-insights WebVTT corpus snapshot generated on 2026-08-25. Academic evidence checked against primary sources through 2026-08-28. arXiv 2605.18747 (Code as Agent Harness) is the primary code-centric survey; arXiv 2606.20683 supplies the six-responsibility runtime taxonomy.