Descriptions:
Ray Amjad covers several Claude Code updates that improve the planning phase of agentic coding workflows. The most significant is the configurable plans directory: by setting a custom path in `.claude/settings.json`, developers can store plans alongside project files, making it easy to `grep` historical plans, tag them for other CLI tools, or load them into a fresh Claude Code session as reference context.
The video’s centrepiece is an automated plan review pipeline built on `PostToolUse` hooks. After Claude Code finishes a planning session, two hooks fire automatically: one archives the plan to a timestamped folder for future reference, and a second passes the plan headlessly to both Codex CLI and Gemini CLI for independent critique. If either reviewer returns a blocking verdict, Claude Code redoes the plan before execution — adding automated quality control with no manual steps required. Amjad shows a live run where an 78,000-token planning session triggers both reviewers and Claude Code then incorporates their feedback into a revised plan.
Additional updates include new planning mode options — “Yes, clear context and auto-accept edits” and “Yes, clear context, manually approve edits” — that start execution in a fresh context window after planning, significantly extending the runway for implementation. MCP tool search is now configurable to kick in automatically when tool descriptions would consume an excessive portion of the context window, avoiding the clutter of loading every MCP server’s full schema upfront. Planning mode also arrives in the Claude Code desktop application, enabling parallel planning agents from a GUI.
📺 Source: Ray Amjad · Published January 18, 2026
🏷️ Format: Tutorial Demo
