Resume Consultant Harness

Resume Consultant Harness

A Claude Code skill that produces a company-tailored resume using a **Planner → Generator → Evaluator** three-agent harness — adapted from Anthropic's *[Harness Design for Long-Running Application Development](https://www.anthropic.com/engineering/harness-design-long-running-apps)* (Prithvi Rajasekaran, 2026).

**[한국어 README](README.ko.md)** · Version: see [VERSION](VERSION) · Changelog: [CHANGELOG.md](CHANGELOG.md)

---

Why This Exists

Ask any AI to "write me a resume for this job" and you get the same output every time:

• Empty adjectives — passionate, results-driven, detail-oriented, team player
• Abstract nouns with no referent — business impact, synergy, organizational capability
• Rule-of-three AI rhythm — "designed, implemented, and deployed" on every bullet
• Hallucinated metrics — "improved efficiency by 37%" with no source
• The generator cheerfully rates its own mediocre output as "great"

Hiring managers detect AI-written resumes in under three seconds and throw them away.

The root cause — identified by Anthropic in the code domain — is **self-evaluation bias**: generators confidently praise their own work on subjective tasks where there is no compiler error or test failure to contradict them. Resumes are *more* subjective than code.

This skill solves it by **separating generation from evaluation** across distinct agent roles, and by adding a dedicated final AI Slop sweep that re-verifies the resume from scratch.

---

Architecture

                      ┌─────────────┐
                      │   Planner   │
                      │  spec.md    │
                      └──────┬──────┘
                             │
             ┌───────────────▼───────────────┐
             │     Sprint Loop  S1 … S7      │
             │                               │
             │  Generator ──► deliverable    │
             │       │                       │
             │       ▼                       │