This release introduces a unified code review and fix workflow that consolidates the review, triage, planning, and execution phases into a single /dev:review command. The workflow spawns specialized reviewers in parallel, filters findings by confidence, plans fixes with proper dependencies, and executes them with automatic GitHub thread resolution.
Other highlights include hunk-based C++ formatting that only applies clang-format to changed lines, Python formatting support via ruff, and a go run fallback for shfmt when the binary isn’t installed.
Breaking change: The dev:addressing-pr-comments skill has been replaced by the @dev:reviewers:github agent. Use the GitHub reviewer agent instead, or run /dev:review on a branch with an open PR. The new dev:addressing-reviews skill provides best practices for responding to review comments and resolving threads.
💥 Breaking Changes
Section titled “💥 Breaking Changes”Refactor code review to integrate GitHub PR feedback
Section titled “Refactor code review to integrate GitHub PR feedback”Jan 23, 2026 · @mavam, @claude
The addressing-pr-comments skill has been replaced by the @dev:reviewers:github agent. This agent is now spawned automatically by /review when a PR exists, extracting unresolved GitHub comments as findings alongside the other code reviewers.
If you previously used the skill directly, use the GitHub reviewer agent instead or run /review on a branch with an open PR.
Other changes:
- The
reviewing-changesskill now auto-creates the review directory via hook - Added
GITprefix for GitHub findings in the issue ID table
🚀 Features
Section titled “🚀 Features”Go run fallback for shfmt in format hook
Section titled “Go run fallback for shfmt in format hook”Jan 28, 2026 · @mavam, @claude
The format hook now falls back to go run for shfmt when the binary is not installed directly. This ensures shell scripts get formatted even when shfmt is not on your system, as long as Go is available. The Go module version is pinned to mvdan.cc/sh/v3/cmd/shfmt@v3.10.0 for reproducibility.
Unified code review and fix workflow
Section titled “Unified code review and fix workflow”Jan 26, 2026 · @mavam, @claude · #12
The code review workflow now consolidates review and fix operations into a unified four-phase process. The /dev:review command serves as the primary entry point, orchestrating a complete workflow that reviews, triages, plans, and executes fixes.
The workflow consists of four phases:
-
Review phase: Spawns seven specialized reviewers in parallel to analyze code changes and generate confidence-scored findings.
-
Triage phase: The
@dev:triageragent filters low-confidence findings (below 70%), groups related findings by root cause, and deduplicates cross-reviewer overlap for higher-quality findings. -
Planning phase: The
@dev:planneragent creates a task list from triaged findings with proper dependencies and ordering by severity and file location. -
Execution phase: The
@dev:fixeragent processes fix tasks with mode-specific behavior. In PR mode, each fix spawns an individual prompt to commit, push, and resolve GitHub threads. In batch mode, all fixes are applied autonomously with a single summary commit at the end.
The workflow supports session resumption—if interrupted during the execution phase, running /dev:review again offers to continue from where you left off.
The /dev:fix command is removed; its functionality is now integrated into the unified review workflow. A new addressing-reviews skill provides GitHub PR communication best practices for handling review comments.
Iterative workflow for addressing review findings
Section titled “Iterative workflow for addressing review findings”Jan 23, 2026 · @mavam, @claude
New /fix command orchestrates an iterative workflow for addressing review findings one by one. Each fix runs in a dedicated @dev:fixer agent (Opus) that commits the change, pushes to the remote, and resolves GitHub threads automatically for GitHub findings (GIT-* IDs).
This restores the iterative “fix, commit, reply, resolve” workflow that was lost when /address-pr-comments was converted to a reviewer agent. The new design generalizes to all reviewer findings—not just GitHub comments—and gives Opus fresh context for each fix without accumulated distractions.
The /review command now offers both /fix and plan mode options in its workflow. Both paths handle automatic resolution of GitHub threads after fixes are pushed.
Hunk-based formatting for C++ files in format hook
Section titled “Hunk-based formatting for C++ files in format hook”Jan 23, 2026 · @mavam, @claude
The format hook now applies clang-format to only the changed hunks in staged C++ files rather than the entire file. This keeps your git history cleaner by preventing unrelated formatting changes from appearing in commits when only a few lines have been modified.
When you stage C++ file changes, the hook detects the modified line ranges using git diff --cached and passes them to clang-format via the --lines option. If no staged changes exist, the hook falls back to formatting the entire file. The implementation uses cross-platform compatible sed and bash features to work reliably on macOS and Linux.
Enhance format hook with Python support and helper function
Section titled “Enhance format hook with Python support and helper function”Jan 23, 2026 · @mavam, @claude
The format hook now supports Python formatting via ruff, running both ruff format for code formatting and ruff check --fix for auto-fixable lint issues.
🔧 Changes
Section titled “🔧 Changes”Improved opencode CLI invocation in plan-reviewer agent
Section titled “Improved opencode CLI invocation in plan-reviewer agent”Jan 23, 2026 · @mavam, @claude
The plan-reviewer agent now uses a heredoc for the opencode CLI prompt, improving readability and making the review criteria visible directly in the agent definition. The --variant flag is conditionally included only when a variant is specified, avoiding empty argument errors.
Plan review skill and agent
Section titled “Plan review skill and agent”Jan 23, 2026 · @mavam, @claude
Plan review functionality has been refactored from hook-based scripts into a structured skill and agent. The new reviewing-plans skill provides a formal methodology for evaluating implementation plans across five weighted dimensions (completeness, correctness, feasibility, risk, and clarity) with consistent severity levels for findings. The plan-reviewer agent exposes this capability as a reusable workflow that integrates with plan mode. Hook-based review triggers have been removed in favor of this more robust, composable approach.
🐞 Bug Fixes
Section titled “🐞 Bug Fixes”Improve branch naming instructions in pr-maker agent
Section titled “Improve branch naming instructions in pr-maker agent”Jan 22, 2026 · @mavam, @claude
The pr-maker agent now provides more explicit branch naming instructions to prevent misinterpretation by the Haiku model. The instructions now clearly state that topic/ is a literal prefix, specify the placeholder as <kebab-case-description> for clarity, and include concrete examples (topic/add-user-auth, topic/fix-login-bug, topic/refactor-api) so the agent consistently creates branches with the correct naming pattern.
📝 Other
Section titled “📝 Other”Reviewer improvements
Section titled “Reviewer improvements”Jan 23, 2026 · @mavam, @claude
The plan reviewer agent has been renamed from @dev:plan-reviewer to @dev:reviewers:plan for naming consistency with other reviewers.
The /dev:review command now selects reviewers intelligently based on the actual diff content. The model reads the diff and decides which reviewers are relevant, skipping those that clearly have nothing to review. The github reviewer is spawned only when a PR exists.