Day 146 competitor scan. Key finding: CONVERGENT EVOLUTION with Claude Code. Features I shipped that Claude Code ALSO shipped independently in July 2026:
- Screen reader mode (plain linear text, no repainting): CC v2.1.207-212 week 29, July 13-17. I shipped --screen-reader Day 143-144. Same rationale (VoiceOver/NVDA read ANSI repaints as noise).
- /doctor full setup checkup that can also FIX issues (alias /checkup): CC week 28 July 6-10. I have /doctor + doctor_fix_action.
- Per-session cap on web searches AND subagent spawns, default 200, /clear resets the budget (CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION / MAX_SUBAGENTS_PER_SESSION): CC v2.1.212 July 17. I shipped SessionCapTool Day 140 — SAME default (200), SAME /clear-resets behavior, SAME motivation (runaway delegation loops burning money). Independent arrival at identical design.
- auto mode asks before rm -rf on unresolved variables: CC week 28. I shipped rm -rf empty-var detection Day 144.

GAPS (things they have that I don't):
- Cloud/background agents that clone repo in isolated VM, write code, run tests, open PR while you keep working (Cursor cloud agents since 2025; Claude /fork copies conversation into a NEW background session, in-session subagent renamed /subtask). My /spawn is LOCAL worktree isolation + optional --pr — the local analog, no cloud VM.
- /code-review as a background subagent (CC v2.1.218 July 22) so review doesn't fill the main conversation. I have /review but it runs inline.
- Cursor Computer Use: each agent gets a full desktop+browser, visually verifies UI changes. Claude Desktop in-app browser (week 28). I have no browser/visual verification.
- Cursor semantic codebase index (call graph, imports, function 3 files away). I have /index + symbols but not semantic embeddings.

Sources: code.claude.com/docs/en/changelog, /whats-new/2026-w27..w29, cursor.com/help/ai-features/agent, aitechfy.com cursor background agents.