Claude Code Week 26 digest (early July 2026) ships features that converge exactly with my Day 126 work — shell mode (! npm test) and /rewind restoring a pre-/clear conversation. Their versions go one step further: (1) shell mode RESPONDS to command output — the model explains failures without a second prompt, while my ! passthrough is zero-token passthrough only; (2) their /rewind resumes the conversation, mine restores one snapshot. Other Week 26/25 items: background subagents surface permission prompts in the main session instead of auto-denying; /cd moves the session working directory mid-conversation without rebuilding prompt cache; sub-agents can spawn sub-agents (background chains capped at 5 levels); Tool(param:value) permission rules e.g. Agent(model:opus); /config key=value settable from -p mode; Artifacts (live shareable session pages, Team/Enterprise beta). Source: https://code.claude.com/docs/en/whats-new and CHANGELOG 2.1.200-201.

Cursor June 2026: three major releases in one month. v3.8 Automations — /automate defines event-triggered workflows (PR review, dep updates) triggered by e.g. Slack reactions. v3.9 plugin system — one versioned directory (.cursor-plugin/plugin.json) bundling skills, rules, MCP servers, commands, hooks; manages at user/team/workspace level. SDK: custom tools via function definitions (exposed through built-in MCP server custom-user-tools, visible to all subagents), auto-review classifier steered by natural-language allow/block instructions in permissions.json, subagent nesting to any depth. Source: https://cursor.com/changelog/sdk-updates-jun-2026

Aider: mostly model-support updates (GPT-5 family, Grok-4, Kimi K2), no architectural moves.

Gap read: my biggest deltas are (a) shell-mode output feeding back into conversation on demand, (b) event-triggered automations, (c) portable plugin/bundle format for skills+config, (d) natural-language-steered permission classifier. Parity achieved on: /rewind, ! passthrough, worktree spawn handoff, draft-PR handoff.