CV Boutique Agency — Claude Code Skill
If you use or adapt this project, a credit or link back to this repository is appreciated — see Citation / credit.
A top-tier boutique resume agency skill for Claude Code. Acts as a panel of eight expert reviewers — AI veteran, HR specialist, founder, business operator, devil's advocate, creative reframer, QC lead, and hallucination detector — to build, critique, rewrite, and quality-control resumes until they reach 8.5+ quality with zero hallucinated claims. Final output includes a send-ready .docx.
What this skill does
Core capability
The skill orchestrates a virtual agency panel that interviews you, drafts your resume section by section, challenges every claim, scores the result using a weighted multi-expert model, and iterates until the resume passes the stop conditions (overall score ≥ 8.5, no critical blocker flags, user acceptance).
Three operating modes
Mode 1 — Build from scratch
Starts with a structured intake (language → target industry → role/seniority → geography/company stage), builds the expert panel, then works through your resume section by section: header, summary, achievements, experience, education, skills. Each section is drafted, QC-reviewed, scored, and refined before moving on.
Mode 2 — Uploaded resume critique
Paste or upload your existing resume. The skill detects it, asks only for missing targeting context, then delivers a full executive verdict: strengths, weaknesses, per-section scores, critical blocker flags, and an exact fix list. You then choose to rewrite, rebuild, or tailor.
Mode 3 — Fast mode
Ask only the minimum blocking questions, run the full agency process behind the scenes, enforce all QA and hallucination controls, and deliver the strongest possible result with minimal back-and-forth.
Expert panel
Mandatory (always active):
- AI veteran
- HR / recruiter specialist
- Founder / entrepreneur
- Business leader / operator
- Devil's advocate
- Brainstorm / creative reframer
- Quality-control lead
- Hallucination detector
Context-dependent additions (spawned by the CEO as context reveals the need):
- Domain expert + industry-specific reviewer (when industry is provided)
- Language / localization expert (when language ≠ English)
- ATS specialist (corporate / large org targets)
- Executive branding expert (director+ / C-suite / partner roles)
When Claude Code agent teams are enabled, each of these experts runs as a real independent Claude instance — own context window, own task, direct inter-expert messaging. See Agent teams architecture below.
Scoring model
Each expert scores independently. A weighted overall score is computed. The skill also maintains internal per-section scores and tracks critical blocker flags. It stops only when all three stop conditions are met: score ≥ 8.5, zero blockers, user acceptance.
| Score range | Meaning |
|---|
| 0 – 3 | Weak / non-competitive |
| 4 – 6 | Usable but weak |
| 7 – 8.4 | Strong with clear issues |
| 8.5 – 8.9 | Send-ready strong |
| 9.0 – 9.4 | Must-call quality |
| 9.5+ | Exceptional (rare) |
Hard rules
- No hallucinated claims — ever
- No unsupported percentages or metrics
- No fake or inflated seniority
- No generic self-praise without evidence
- No sugarcoating — if something is weak, the panel will say so
- Memory (
workspace/memory.json) and progress (workspace/progress.json) are updated at every stage
Final output bundle
- Final resume (formatted text)
- Panel report (per-expert verdicts)
- Scorecard (per-section + weighted overall)
- Rewrite / action plan
- Downloadable
.docx
Optional extras: tailored variants, LinkedIn summary, cover letter starter, interview talking points
Fun epilogue: once your resume is delivered, the agency offers to channel a famous figure from your industry (or a historical wildcard) for a 3–4 sentence in-character reaction to your resume — Elon Musk, Warren Buffett, Marie Curie, Sun Tzu, and more. Purely for fun; zero effect on the resume or scores.
Repository layout
stevesolun