This release improves workflow automation for changelog creation and code review. The changelog-adder agent now auto-detects changed files and provides clearer guidance for writing user-focused entries. The review command streamlines decision points, reducing prompts while maintaining control over important choices.
🔧 Changes
Section titled “🔧 Changes”Automated scope detection and improved changelog guidance in changelog-adder agent
Section titled “Automated scope detection and improved changelog guidance in changelog-adder agent”Jan 30, 2026 · @mavam, @claude
The changelog-adder agent now automatically injects context about changed files through a PreToolUse hook, eliminating manual scope detection. The hook runs an improved scope detection script that prioritizes staged changes, then unstaged changes, then branch history.
The agent’s instructions have been refined to provide clearer guidance for writing user-facing changelog entries. New sections explicitly distinguish what to exclude (internal implementation details), what to preserve (user-facing terms), and what to include (outcomes and examples). A self-review checklist helps ensure changelog entries are clear and helpful.
These improvements streamline the changelog creation workflow by automatically providing essential context while guiding the agent toward writing changelogs that focus on user benefits rather than implementation details.
Streamlined review workflow with structured decision points
Section titled “Streamlined review workflow with structured decision points”Jan 29, 2026 · @mavam, @claude
The /dev:review command now streamlines the workflow by using Claude’s structured AskUserQuestion tool to consolidate decision points into focused, meaningful interactions.
Phase 2 (Triage) now presents structured options after showing findings:
- Fix all findings (Recommended)
- Select which findings to skip
- Abort review
Phase 3 (Plan) proceeds automatically after the user approves findings in Phase 2, eliminating the intermediate plan review prompt.
Phase 4 (Execute) optimizes prompting based on finding type:
- Findings from automated reviewers (security, architecture, tests, and others) execute without per-task prompts
- GitHub PR comments only trigger an additional prompt if they exist, with options to address all, review each comment individually, or skip
- Per-finding prompts only appear when the user explicitly selects “Review each comment” for GitHub feedback
This reduction in prompts—from 2+ decisions down to 1 for automated reviewers, and conditionally 2 for workflows with GitHub comments—keeps the review workflow responsive while maintaining full control over important decisions.