Develop Honest Resumes Through Adversarial Feedback

Develops resumes that are both compelling and factually accurate by orchestrating four specialized agents through repeated cycles of feedback. The Writer creates persuasive framing, the Fact-Checker catches hallucinations before they spread, the Interviewer critiques like a real hiring manager would, and the Coach synthesizes all feedback into actionable guidance. This prevents the two common resume failures: underselling your genuine achievements or inflating your experience with invented details.

You provide your raw experience and optionally a target job description. Run /resume-helper:resume-loop with your files and optional page limit. The system orchestrates an iterative loop where each iteration runs four agents in sequence: 1. Writer drafts a resume using only information from your files. 2. Fact-Checker traces every claim back to your input, rejecting fabrications. 3. Interviewer reviews skeptically from a hiring manager's perspective (seeing only the resume, not your raw background). 4. Coach reads all outputs, runs parallel analysis agents for vague claims, buzzword detection, and ATS scoring, then issues a verdict and asks clarifying questions. You answer questions between iterations. The loop repeats until Coach issues READY. Output: resume_final.md, interview_prep.md, and assessment.md audit trail.

Requires Claude Code plugin environment with local file write access. Agents use Sonnet by default; --premium flag upgrades Coach to Opus for higher-quality synthesis at higher cost. The plugin.json file defines the command structure and agent metadata. File-based communication avoids API round-trip overhead.