Claude Code v2.1.154: Opus 4.8, Dynamic Workflows, and Major Fixes
Anthropic released Claude Code v2.1.154, adding Claude Opus 4.8 and dynamic workflows. The model defaults to high effort; use /effort xhigh for the hardest tasks. Fast mode is available at 2x the standard rate for 2.5x speed — a significant cost reduction.
Dynamic Workflows
The headline feature: you can now ask Claude to create a workflow, and it orchestrates work across tens to hundreds of background agents for larger, more complex tasks. Use /workflows to view running workflows.
Agent & Shell Enhancements
In claude agents, prefix a shell command with ! to run it as a background session you can attach/detach from. The equivalent CLI flag is claude --bg --exec '<command>'. Also, /logout now actually signs you out (not sent to a background session). The agents view (double left arrow) now works on Bedrock, Vertex, Foundry, and with telemetry disabled.
Plugin Changes
Plugins can now declare defaultEnabled: false in plugin.json or a marketplace entry. Enable them with /plugin or claude plugin enable; dependencies remain auto-enabled. The Discover tab pins plugins matching your current directory with a “suggested for this directory” annotation.
Other Notable Improvements
- Lean system prompt is now default for all models except Haiku, Sonnet, and Opus 4.7 and earlier.
- Multiple-choice prompts are reserved for decisions Claude genuinely cannot make.
/simplifynow runs a cleanup-only review (reuse, simplification, efficiency, altitude) and applies fixes — no longer runs full bug-hunting/code-review --fix.- Streaming tool execution is always enabled, including on Bedrock/Vertex/Foundry and with telemetry disabled.
- Stdio MCP server subprocesses receive
CLAUDE_CODE_SESSION_IDandCLAUDECODE=1environment variables. /chromelets you pick which connected browser to use when multiple are connected.
Bug Fixes
- Fixed
rm -rf $HOMEnot being blocked whenHOMEhas a trailing slash. - Fixed
$TMPDIRresolving inconsistently between sandboxed and unsandboxed Bash commands. - Fixed pinned background sessions respawning every minute after a Claude Code update, causing notification churn.
- Fixed subagents in background sessions bypassing worktree-isolation guard and writing to shared checkout.
- Fixed orphaned
claude --bg-pty-hostprocesses spinning at 100% CPU on macOS. - Fixed
worktree.baseRef: "head"resolving to main checkout’s HEAD instead of current worktree’s HEAD. - Fixed intermittent terminal rendering corruption in VS Code (capped spinner colors).
- Many others, including background session state management, keyboard shortcuts, and classifier detection of data exfiltration.
📖 Read the full source: GitHub Claude-Code
👀 See Also
Coinbase x402 vs Google A2A: Two Opposite Payment Orderings for Agent-to-Agent Payments
Building agent-to-agent payments reveals a fundamental split: Coinbase's x402 middleware settles after work (verify→run→settle), while Google's A2A extension settles before (verify→settle→run) for slow agentic calls.
Anthropic study reveals cognitive degradation in AI-assisted workflows
Anthropic's global study of 80,000 users found academic users report cognitive degradation rates 2.5x higher than average when using AI tools like Claude and Cursor. The source identifies the problem as users eliminating the 'digestion phase' of work.
DeepSeek-V4-Flash Makes LLM Steering Practical for Local Models
Seen Goedecke explains why steering vectors are relevant again thanks to DeepSeek-V4-Flash running locally via DwarfStar, with hands-on details on how steering works and why it hasn't caught on before.
Claude Agent SDK Billing Changes June 15: Per-User Credits, No Rollover, Hard Cliff
Starting June 15, Claude Agent SDK usage and claude -p stop counting against subscription limits. Each user gets a separate monthly credit (e.g., Pro $20, Max 5x $100). Credits don't pool, don't roll over, and have a hard cliff.