Skip to content

Thursday, March 26, 2026

Daily picks

25

articles scored

#1 GOLDReleaseClaude Code Releases

v2.1.84

  • Windows users get a PowerShell tool as an opt-in preview — finally a native shell experience instead of relying on bash workarounds
  • If you've ever come back to a stale Claude Code session hours later and burned tokens re-caching context, a new idle-return prompt now nudges you to /clear after 75+ minutes away
  • New TaskCreated hook fires when a task spins up via TaskCreate — useful for logging, notifications, or triggering side effects in multi-agent workflows
  • WorktreeCreate HTTP hooks can now return the created worktree path, making it easier to integrate with external CI systems
  • MCP tool descriptions are now capped at 2KB, preventing OpenAPI-generated servers from stuffing your context with thousands of tokens of schema noise
  • Token counts ≥1M now display as '1.5m' instead of '1512.6k' — small detail, big improvement for readability in long agentic sessions
#2 SILVERReleaseGitHub anthropics/claude-code-action

Claude Code GitHub Action v1.0 — Now Generally Available

  • The official Claude Code GitHub Action hits v1.0 GA — it's now the stable, supported way to wire Claude into your CI/CD pipelines
  • Mode detection is now automatic: no more manually setting 'interactive' vs 'automation'. The action figures it out from your setup
  • All prompts go through a single 'prompt' input now, and CLI options through 'claude_args' — simpler, more consistent with how you'd use the CLI directly
  • Ready-made examples cover automated PR review, CI failure fixes, issue triage, doc generation, and security scanning
  • Breaking changes from v0.x: if you're already using this action, check the migration guide before updating
#3 BRONZEReleaseGitHub anthropics/claude-agent-sdk-typescript

[Release] anthropics/claude-agent-sdk-typescript: v0.2.84

  • New 'taskBudget' option lets you tell the model upfront how many tokens it has for a task, so it paces tool use instead of going all-in on exploration and hitting limits mid-task
  • 'enableChannel()' method and 'capabilities' field on McpServerStatus give you SDK-level control over which MCP channels are active

Made with passive-aggressive love by manoga.digital. Powered by Claude.