feat(skills): add eval-harness skill and eval command

[baleen37]

Summary

• Add skills/eval-harness/SKILL.md — eval-driven development (EDD) skill with capability/regression eval types, pass@k/pass^k metrics, code/model/human grader types, and define→implement→evaluate→report workflow
• Add commands/eval.md — /eval slash command with define, check, report, list, and clean subcommands for managing eval definitions in .claude/evals/

Test plan

• Verify /eval command triggers correctly in Claude Code session
• Verify eval-harness skill loads when EDD-related queries are asked
• Test /eval define <name> creates eval definition file
• Test /eval list shows existing evals
• Verify SKILL.md frontmatter follows plugin skill conventions (third-person description, trigger phrases)

Summary by CodeRabbit

Release Notes

• Documentation
  • Added comprehensive Eval Harness documentation covering Evaluation-Driven Development (EDD) concepts, eval types, grading methods, metrics, and workflow guidance.
  • Added Eval command documentation detailing subcommands for defining, checking, reporting, listing, and managing evaluations with examples.

[coderabbitai]

Caution

Review failed

The pull request is closed.

📝 Walkthrough

Walkthrough

This pull request adds two documentation files establishing eval-driven development practices: commands/eval.md provides a command reference guide with subcommands and workflows, while skills/eval-harness/SKILL.md offers comprehensive documentation on evaluation-driven development methodology, grading strategies, and implementation examples.

Changes

Cohort / File(s)	Summary
Eval Command Documentation commands/eval.md	Introduces the eval command with subcommands (define, check, report, list, clean), including argument parsing, directory structure under .claude/evals/, and examples of capability and regression evaluations.
Eval Harness Skill Documentation skills/eval-harness/SKILL.md	Comprehensive guide on Evaluation-Driven Development (EDD) covering methodology, eval types (Capability/Regression), grading options (Code-Based/Model-Based/Human), metrics (pass@k, pass^k), workflow stages, storage conventions, and detailed examples including authentication eval implementation.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐰 A harness for evals, so clever and bright,
Commands laid out for the developer's night,
Define, check, and report with such care,
Testing AI's wit—oh what a fair pair!
Evaluation-driven, we hop toward the light! 🌟

✨ Finishing touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch feat/eval-harness-skill

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}