v1.3.0 — Socratic Code Review Coaching

Think Before Code now reviews code without stealing the review from you.

v1.3.0 introduces code-review-coach, the suite’s first deliberately non-DSA skill. It helps learners practise reviewing existing code, diffs, and pull requests through guided questioning—one concern at a time—instead of receiving a dumped findings list or a rewritten implementation.

Highlights

• Added code-review-coach, bringing the suite to nine skills.
• Establishes the code’s contract before judging it.
• Requires concrete evidence before impact, and impact before severity.
• Guides the learner toward the smallest justified change and one verification idea per finding.
• Prevents premature design-pattern prescriptions and “best practice” cargo culting.
• Ends with a prioritized review summary written by the learner.
• Hands concrete failures to debug-coach, systematic suite design to test-case-coach, and complexity-only questions to complexity-coach.

Evals and examples

• Added 15 activation and boundary rows for the new review workflow.
• Added seven dedicated code-review-coach behavior cases.
• Added three cross-skill boundary cases covering review vs. debugging, test design, and complexity analysis.
• Added a complete non-DSA code-review transcript demonstrating evidence-backed findings, prioritization, a declined Strategy pattern, and a learner-authored final review.

Repository improvements

• Updated the README, decision guide, repository tree, roadmap, and contributing guidance for the nine-skill suite.
• Updated the interactive demo’s displayed skill list.
• Strengthened neighboring skill boundaries around code review.
• Updated the structural validator’s expected skill set.
• Refreshed GitHub Actions dependencies.

The central rule remains unchanged:

One hint at a time. Think before code.

v1.2.0 — Test-Case and Pattern-Transfer Coaching

What’s new

Think Before Code now includes two new specialist Agent Skills:

• test-case-coach — helps learners design compact, justified test suites through boundaries, adversarial inputs, expected outputs, and regression cases.
• pattern-transfer-coach — helps learners extract reusable structure from solved problems, identify recognition and rule-out signals, and adapt the idea to one cousin problem.

Also improved

• Expanded the suite from six to eight skills
• Added two realistic example transcripts
• Added 24 activation-prompt cases and 12 behavior cases
• Strengthened cross-skill routing boundaries
• Added positive and negative activation coverage checks for every skill
• Updated the README, roadmap, repository tree, and validation documentation

Validation

OK: 8 skill(s) validated, no problems found.
OK: 72 activation-prompt rows validated; every skill has positive and negative activation coverage.

v1.1.0 — Proof of Learning

Think Before Code v1.1.0 — Proof of Learning

This release expands Think Before Code from a collection of Socratic tutoring skills into a more testable and installable learning system.

Added

• debug-coach, a dedicated skill for debugging learner-written code without rewriting the solution
• A complete end-to-end tutoring transcript
• 48 activation and boundary-test prompts across all skills
• Detailed behavioral evaluation cases
• A dependency-free structural validator for Agent Skills
• GitHub Actions validation for pushes and pull requests
• Cross-agent installation instructions for Claude Code, VS Code, GitHub Copilot, and Codex
• A project changelog

Improved

• Updated the roadmap to reflect completed work
• Clarified boundaries between the specialist skills
• Expanded mistake-log documentation
• Improved Quick Start and installation guidance
• Strengthened skill frontmatter and structural validation

Current skills

• dsa-tutor
• problem-decoder
• dry-run-coach
• complexity-coach
• mock-interviewer
• debug-coach

Quackrates now has CI, evaluations, versioning, and considerably more authority than a rubber duck should possess. 🦆