Evaluate Candidates Consistently Across Your Hiring Pipeline

Runs resume evaluations with consistent rubrics and eliminates bias from hiring by filtering presentation style from technical substance. It applies the same scoring criteria to every candidate regardless of when their application arrives, removing fatigue-based inconsistencies and helping you surface overlooked potential that gut-feel hiring misses.

Once you set up your organization's role levels, rubrics, and tech stack, you use the framework in stages: 1. Run /evaluate-resumes to score all new resumes against your IC or Manager rubric, auto-generating HR summaries. 2. For promising candidates (✅ and ⭐ scores), run interview prep to generate role-targeted questions. 3. Fill in interview notes during the interview, then run /assess-interviews to score responses. 4. Use the candidate-comparer to side-by-side finalists. Output: structured evaluations with breakdown scores, interview assessments, and batch reports with CSV tracking.

Every candidate in a hiring pipeline gets evaluated on the same substantive criteria with the same rigor, so great potential isn't lost to bias or reviewer fatigue, and your final hire decision is backed by structured evidence instead of impression.

Requires either Claude Code CLI or VS Code with GitHub Copilot enabled. The shared methodology lives in AI.md (single source of truth for both providers), while CLAUDE.md and .github/copilot-instructions.md contain provider-specific dispatch logic. All candidate data is gitignored (.org/*, resumes/, evaluations/, reports/), keeping sensitive hiring information out of version control while keeping templates and agent definitions committed.

This framework is intentionally structured and opinionated. If your hiring process is highly informal or non-technical roles, the 7-10 point rubrics may feel overly granular. Setup requires configuring role levels and rubrics for your organization (one-time step). The regional context features only activate for candidates from regions you explicitly configure.