WHY you should care
Two API-level additions in v2.1.218 affect how you build with Managed Agents in production:
-
effort parameter on agent create — `BetaManagedAgentsModelConfigParams` now accepts `effort: "low" | "medium" | "high" | "xhigh" | "max"`. Control reasoning depth per-agent at creation time. Useful for cost/latency tuning.
-
`dreaming-2026-04-21` beta header — New accepted beta on agent create/update. What it enables is unclear yet, but it's live. Worth watching.
Other notable changes
- 🚀 `/code-review` runs as a background subagent — no longer floods conversation context
- 🐛 Windows path fix — `C:\Users...` paths with `\u`-prefixed segments were silently corrupting to CJK characters
- 🐛 Left-arrow no longer silently discards conversations
- `/deep-research` won't auto-launch anymore — manual only
- `context: fork` skills background by default
Source PR
thevibeworks/claude-code-docs#1084
/cc @lroolle
WHY you should care
Two API-level additions in v2.1.218 affect how you build with Managed Agents in production:
effortparameter on agent create — `BetaManagedAgentsModelConfigParams` now accepts `effort: "low" | "medium" | "high" | "xhigh" | "max"`. Control reasoning depth per-agent at creation time. Useful for cost/latency tuning.`dreaming-2026-04-21` beta header — New accepted beta on agent create/update. What it enables is unclear yet, but it's live. Worth watching.
Other notable changes
Source PR
thevibeworks/claude-code-docs#1084
/cc @lroolle