C9PG
Claude Code Docs

Release Changes

Last verified: April 1, 2026 — Claude Code v2.1.89 · Opus 4.6 · Superpowers v5.0.6

Corrections to the original guide

The expanded vibe-coding guide starts by correcting a few important assumptions:

  1. /plan is native Claude Code, not a Superpowers feature. Superpowers adds a methodology layer on top of planning.
  2. The Superpowers command transition is complete in v5.0.6. The current explicit commands are /superpowers:brainstorm, /superpowers:write-plan, and /superpowers:execute-plan. The old slash commands are deprecated redirects.
  3. The hook model is larger than the smaller source document implies. As of March 2026 there are 21+ lifecycle events across 4 handler types.
  4. Several modern workflow features were missing from the original guide, including /rc, /fast, /loop, /batch, /simplify, /debug, /effort, /voice, /color, /rewind, Agent Teams, Auto Mode, Computer Use, --worktree, --from-pr, and the 1M-token context window for Max/Team/Enterprise.
  5. opusplan is still valid, but /fast is a separate speed-oriented toggle for Opus 4.6.
  6. The hook if field added in v2.1.85 remains the recommended way to reduce hook overhead.
  7. defer in v2.1.89 is critical for CI/CD pipelines that need human review on destructive operations.

What Changed Recently

DateVersionChange
Apr 1v2.1.89defer PreToolUse decision, PermissionDenied hook, named subagents in @-mention, thinking summaries off by default
Mar 29v2.1.88Session ID header for proxies, Jujutsu/Sapling VCS support
Mar 24—Auto Mode launched (AI-classified permissions)
Mar 23—Computer Use added to Claude Code and Cowork
Mar 13v2.1.751M token context window default for Opus 4.6 (Max/Team/Enterprise), /color command
Mar 11v2.1.76/effort command, MCP elicitation, Elicitation and ElicitationResult hooks
Mar 7v2.1.71/loop command for recurring tasks
Mar 3v2.1.69/voice push-to-talk mode
Feb 28v2.1.63/simplify and /batch bundled skills
Feb 25—Remote Control (/rc) research preview
Feb 5v2.1.32Agent Teams research preview

Layer 1: Foundation

  • write a tight CLAUDE.md
  • configure settings.json
  • learn Shift+Tab, Ctrl+C, @file, and !cmd
  • run /init, /terminal-setup, and /doctor

Layer 2: Verification

  • add a PreToolUse hook that blocks source edits without RED
  • add PostToolUse formatting and related-test hooks
  • add a Stop verification gate
  • add scoped rules
  • create the test-forward skill

Layer 3: Productivity

  • install Superpowers
  • install LSP plugins
  • use /simplify and /debug
  • enable /rc
  • use /fast when interactive iteration speed matters

Layer 4: Scale

  • use /batch
  • use subagents for writer/reviewer splits
  • experiment with Agent Teams
  • integrate headless mode plus defer into CI/CD
  • use /loop and --worktree where appropriate

Layer 5: Continuous Improvement

  • run /insights monthly
  • review /release-notes weekly
  • codify successful workflows into skills
  • share hooks, rules, and skills through committed team config
  • keep personal overrides in settings.local.json