Claude Code v2.1.239: Cost Estimates Include 1.1x Premium, /claude-api Upgrade, and Major Fixes
Claude Code v2.1.239 (latest) rolls out cost-estimate updates, a new migration command, and a long list of fixes for Bedrock, MCP, and terminal rendering. Here's what's practical for daily use.
Cost Estimates Now Include US-Only Premium
If you're in a data-residency workspace, the /cost command, status line, and --max-budget-usd now factor in the 1.1× premium for US-only inference. Budgeting gets more accurate—no more surprise overruns.
New /claude-api Upgrade Command
For Python projects stuck on the anthropic 0.x SDK, /claude-api upgrade migrates them to 1.x. The skill's reference now uses anthropic.Timeout instead of httpx.Timeout—check your timeout calls after upgrading.
Fullscreen Renderer Opens Up
The one-time fullscreen renderer offer is now available on Bedrock, Vertex, Foundry, and other previously excluded setups. New installs there start in fullscreen (if you accept the offer).
Cloud Sessions: Plugin Syncing
Plugins synced from claude.ai now show as name@synced and work with claude plugin enable/disable <name>@synced. They won't override a plugin you installed manually with the same name.
Key Fixes
- Bedrock streaming: Fixed double-billing when proxies strip Content-Type—now streams properly instead of re-running turns non-streaming.
- HTTPS proxy hang: Fixed startup hang with Bedrock SSO profiles and
awsAuthRefresh; credential pre-check now honorsHTTPS_PROXY. - JetBrains IDE: Fixed 5-second pause on Edit/Write when the plugin is connected.
- WebFetch memory: Expired page content now clears after 15 minutes instead of persisting all session.
- Vim mode: Escape in agent view now exits to NORMAL without clearing the prompt.
- MCP forms: Tall elicitation forms now fit the fullscreen window, with scrolling and visible Accept/Decline.
- Remote MCP: Servers recover after transient 5xx on reconnect in cloud sessions or via
setMcpServers(). - Themes:
dark-ansino longer renders expanded tool results with invisible text. - Fullscreen prompt: Stops reappearing after being shown three times if it can't be answered.
- Path matching:
/resumeno longer picks up sessions from similarly-named dirs (e.g., underscores vs. dashes). - UTF-8 BOM: Agents, skills, and commands with BOM in .md files are now recognized.
- OpenTelemetry: Deferred tool executions resume in the original trace, not a new one.
- Mouse reports: Fixed terminal inserting escape sequences like
35;150;7Mwhen reports split across writes.
Who It's For
If you use Claude Code on Bedrock, in JetBrains, or rely on cost tracking for data-residency, this release directly addresses your pain points. The /claude-api upgrade is essential if you're still on the Python 0.x SDK.
📖 Read the full source: GitHub Claude-Code
👀 See Also

MCP Server Enables AI Agents to Make Real Purchases with Ephemeral Virtual Cards
A developer has built an MCP server that allows AI agents to complete real purchases using ephemeral virtual Visa cards issued just-in-time. The system requires user approval via MFA and issues cards locked to specific merchants with 15-minute TTLs.

Fullerenes: Open-source persistent memory layer for coding agents cuts tokens by 64% on SWE-bench
Fullerenes uses a local SQLite knowledge graph built via Tree-sitter to give coding agents like Claude Code persistent memory, reducing token usage by 64% on SWE-bench and up to 96.6% on internal benchmarks.

OpenClaw Optimizer v1.18.0 released with OpenClaw v2026.3.7 alignment
OpenClaw Optimizer skill v1.18.0 is now aligned with OpenClaw v2026.3.7, adding support for new AI providers including Google Gemini 3.1 Flash-Lite and OpenAI gpt-5.4, plus new CLI commands like /session idle and /usage cost.

OpenClaw user creates 'feelslikeclaude' skill to improve ChatGPT agent workflow behavior
A developer switched their OpenClaw setup from Claude to ChatGPT and found the key difference was workflow behavior, not writing style. They created a clawhub skill called 'feelslikeclaude' to nudge ChatGPT toward better execution habits.