Develop Honest Resumes Through Adversarial Feedback14 skills
A multi-agent iterative loop that pairs resume writing with fact-checking, skeptical review, and career coaching.
Career Blueprint
Safety checkedDevelop Honest Resumes Through Adversarial Feedback
MadeByTokensMay 2026
A Claude Code plugin that pairs resume writing with multi-agent fact-checking and skeptical review to prevent both underselling and overselling.
Resume WritingFact ValidationSkeptical ReviewCareer CoachingInterview Prep
Install Command
Run this command to deploy the blueprint to your environment.
What problem does this solve?
Develops resumes that balance compelling advocacy with factual accuracy using four specialized agents. The Writer creates compelling framing, the Fact-Checker catches hallucinations, the Interviewer reviews skeptically like a real hiring manager, and the Coach mediates between them. This adversarial approach stops resumes from either underselling qualified candidates or overstating their experience with invented details.
How does it work?
Once installed, you run the main command with your experience file and optionally a job description. The system runs an iterative loop: 1. Writer creates a resume draft using only information you provided. 2. Fact-Checker verifies every claim against your input, rejecting fabrications. 3. Interviewer reviews the draft skeptically (seeing only the resume, not your raw input, just like real hiring managers do). 4. Coach synthesizes all feedback, asks clarifying questions to strengthen weak claims, and issues a verdict (READY, NEEDS_STRENGTHENING, NEEDS_GROUNDING, or BLOCKED). The loop repeats until the Coach judges the resume interview-ready. Output: a polished resume, interview preparation guide, and full audit trail of the development process.
Included Skill Groups
What's the biggest win?
Every claim on your resume stays factually grounded while being presented in the strongest honest light, so you defend every bullet point in interviews instead of apologizing for exaggerations.
What should I know technically?
Requires Claude Code with plugin support and write access to a working/ directory for file-based agent communication. Uses Sonnet by default for all agents except optional --premium flag, which upgrades the Coach agent to Opus for higher-quality synthesis at higher token cost. The generate-premium-agents.sh script auto-generates the premium variant from the base coach.md file.
What should I watch out for?
The loop requires you to answer Coach questions with specific details (team sizes, metrics, timelines, business impact) for the system to strengthen claims; if you provide only vague raw input, you get a vague honest resume rather than a polished strong one. Requires Claude Code environment (not standalone Chat). Best results with 1-2 page resumes; longer backgrounds (10+ years, executive/academic) may need --max-pages 2 or 3 to avoid truncation.
Key Features
Adversarial agent loop
Writer advocates, Fact-Checker validates, Interviewer skeptics, Coach mediates—four perspectives iterate until consensus on readiness
Information isolation
Interviewer sees only the resume (not your raw input), simulating how real hiring managers evaluate without full context
Hallucination prevention
Fact-Checker traces every number, achievement, and claim back to your original input; invented details get rejected before they spread
Parallel analysis agents
Vague claim detection, buzzword identification, ATS keyword matching, and quantification suggestions run simultaneously to provide objective metrics
File-based architecture
Agents communicate through disk files instead of context windows, allowing long development loops without exhaustion
Tools in this Blueprint
About This Blueprint
- License
- MIT
- Industry
- Technology
- Skills
- 1 skills total