⚡
★ 86.6k

Chapter 11: Changelog

Complete release history for Superpowers. Releases are listed in reverse chronological order (most recent first).

v5.0.2 — March 2026

Theme: Infrastructure resilience and session management

What's New

  • Zero-dependency brainstorm server. The brainstorming companion now runs without any external dependencies. No npm install required for the brainstorm server — it runs on Node.js built-ins only. This eliminates an entire class of setup failures and makes the brainstorm feature available immediately after clone.

  • 30-minute idle auto-exit. Long-running sessions that have been idle for 30 minutes automatically exit cleanly. This prevents orphaned processes, reduces resource consumption, and ensures that sessions do not accumulate stale context from hours-old work.

  • Context isolation improvements. Sub-agent contexts are now more strictly isolated from parent contexts. Information from one agent's investigation does not leak into another agent's reasoning, preventing a class of subtle contamination bugs where an agent would "remember" something from a parallel workstream that it should not have access to.

Bug Fixes

  • Fixed an edge case where brainstorm server would fail to start if the default port was in use
  • Corrected context serialization for deeply nested agent trees

v5.0.1 — March 2026

Theme: Platform expansion

What's New

  • Gemini CLI extension support. Superpowers is now fully supported on Google's Gemini CLI through the extension system. All core skills — TDD, systematic debugging, verification, code review, worktrees — are available on Gemini CLI with the same enforcement guarantees as Claude Code. Configuration instructions are included in the installation documentation.

Notes

This is a platform support release. No behavioral changes to existing skills or protocols.

v5.0.0 — March 2026

Theme: Visual reasoning and document intelligence

What's New

  • Visual brainstorming companion. A new companion mode renders the brainstorming process as an interactive visual diagram. See ideas, connections, and decision trees visually as the brainstorm progresses. Particularly useful for complex architecture decisions and system design work.

  • Document review system. New capability for reviewing technical documents — RFCs, architecture decision records, design documents, and specifications — with the same rigor applied to code review. The document review system checks for internal consistency, missing edge cases, and unstated assumptions.

  • Architecture guidance skill. A new dedicated skill for architecture-level decisions. Activates when designing systems, choosing data stores, planning service boundaries, or making decisions that affect multiple components. Provides structured frameworks for trade-off analysis.

Breaking Changes

  • The brainstorm hard-gate from v4.3.0 now also applies to architecture discussions. You cannot begin architecture implementation without completing a brainstorm + document review cycle.

v4.3.1 — February 2026

Theme: Official Cursor support

What's New

  • Cursor IDE official support. Full Superpowers support for the Cursor IDE. All skills, protocols, and Iron Laws are available through Cursor's rules system. Installation documentation covers Cursor-specific configuration, including how to map skills to Cursor's composer and chat interfaces.

  • Cursor-specific workaround for context window limitations in long sessions

  • Dedicated Cursor configuration file format documented

v4.3.0 — February 2026

Theme: Brainstorming enforcement

What's New

  • Brainstorming hard-gate. Creative work — building new features, designing new components, adding new behavior — now requires a completed brainstorm session before any implementation may begin. This is a hard gate: the agent will not proceed to implementation without brainstorm completion. Previously this was a soft recommendation; it is now enforced.

  • Brainstorm session minimum requirements specified: problem exploration, at least two alternative approaches considered, explicit decision made with reasoning documented.

Rationale

Analysis of session logs from v4.2.x showed that 73% of significant rework was on features where brainstorming was skipped. The soft recommendation was not being followed consistently. The hard gate addresses this.

v4.2.0 — February 2026

Theme: Isolation and main branch protection

What's New

  • Mandatory worktree isolation. All feature development now requires a git worktree. Working directly in the main repository checkout on a feature branch is no longer permitted. Every feature, every bugfix, every experiment gets its own isolated worktree directory.

  • Main branch protection. Direct commits to main are now explicitly prohibited by the protocol. All changes must go through a branch and either a local merge (with tests) or a pull request.

  • Updated Chapter 8 with the five-step worktree creation process and four completion options.

Impact

Projects migrating from v4.1.x need to adopt the worktree workflow. The transition guide is in the installation documentation.

v4.0.0 — December 2025

Theme: Review infrastructure and debugging toolchain

What's New

  • Two-stage code review. Code review is now a two-stage process: automated static analysis in stage one, subagent review in stage two. This ensures machine-detectable issues are caught before consuming the subagent's attention, which is reserved for architectural and logical issues.

  • DOT flowcharts in planning. Implementation plans can now include DOT language flowcharts that are rendered visually. Complex multi-step plans with conditional branches are significantly easier to review and understand when visualized.

  • Bundled debugging tools. A set of debugging utilities is now included with Superpowers: structured log analysis, test failure pattern detection, and regression bisection helpers. These tools support the four-phase debugging process described in Chapter 6.

Breaking Changes

  • Plan format updated to support DOT flowchart blocks. Existing plans in the old format continue to work but will not render visual flowcharts.
  • Two-stage review changes the structure of superpowers:requesting-code-review. Update any custom review automation.

v3.5.0

Theme: OpenCode platform support

What's New

  • OpenCode support. Superpowers now supports the OpenCode platform. Configuration and skill mapping documentation included.

v3.3.0 — October 2025

Theme: Experimental Codex integration

What's New

  • Experimental Codex support. Early access support for OpenAI Codex. Core skills are available; some advanced features are limited. Marked as experimental — behavior may change in subsequent releases.

  • Known limitations with Codex: context window constraints affect long planning sessions; some skill triggers require explicit naming rather than automatic detection.

v3.2.0 — October 2025

Theme: Design documentation and namespace

What's New

  • Design docs in brainstorming. The brainstorming workflow now produces a structured design document as output. This document captures the explored alternatives, the decision made, and the reasoning — and feeds directly into the implementation plan phase.

  • superpowers: namespace. All Superpowers skills are now prefixed with the superpowers: namespace (e.g., superpowers:systematic-debugging, superpowers:verification-before-completion). This prevents naming conflicts with user-created skills and makes skill invocations unambiguous.

Migration

Skills from v3.1.x without the namespace prefix continue to work through aliases. The namespace prefix is recommended for all new invocations.

v3.0.1 — October 2025

Theme: Skills system migration

What's New

  • Anthropic first-party skills system migration. Skills are now managed through Anthropic's official first-party skills system rather than the previous custom mechanism. This brings improved reliability, better version management, and access to Anthropic's skill distribution infrastructure.

  • All existing skills migrated to the new system with backward compatibility maintained.

  • New skill installation process documented in the setup guide.

v2.0 — October 2025

Theme: Skills repository separation and expansion

What's New

  • Skills repository separation. The skills library is now maintained as a separate repository from the core Superpowers configuration. This allows skills to be updated independently of the core protocol, and allows teams to maintain private skill repositories alongside the public library.

  • 9 new skills added:

    • systematic-debugging — The four-phase debugging protocol
    • verification-before-completion — The 5-step verification gate
    • receiving-code-review — Technical rigor in review response
    • requesting-code-review — Structured review request protocol
    • using-git-worktrees — Worktree isolation workflow
    • writing-plans — Pre-implementation planning
    • executing-plans — Plan execution with checkpoints
    • dispatching-parallel-agents — Multi-agent coordination
    • writing-skills — Skill creation and testing

Notes

v2.0 marks the point at which Superpowers became a complete development methodology rather than a collection of individual tools. The nine skills in this release cover the full lifecycle of a feature from planning through verification.

Earlier Versions

Versions prior to v2.0 are documented in the legacy changelog available in the repository history. The v2.0 release represented a significant architectural restructuring; earlier version history is retained for reference but the protocols described there have been superseded.