Six weeks ago I wrote about rebuilding my CV from scratch — master file, AI-NOTE annotations, tailored output generated from a job posting URL in under two minutes.
Several people asked: then what?
Then you get put on camera.
The tailored CV ships. Then something else follows — a recruiter screen, an async video submission, a hiring manager who wants sixty seconds in your own voice. Why you, why this role, why now.
Most people default to one of two failure modes. They read the CV out loud — which the recruiter already has — or they hand it to an AI and send a script that’s fluent, polished, and indistinguishable from forty other submissions that week.
Neither works. Both avoid the actual problem: under pressure, most people can’t clearly articulate why they actually want the role.
The CV can’t answer that. You have to.
Two New Layers
The tailoring command now does two things after generating the CV.
Gap tracking. Every application appends a structured entry to a file called gaps.md. Role, company, date, and a table: each gap identified during tailoring, a severity label — hard required, soft required, or soft differentiator — and a one-line note. No softening. If the job description asked for something the master CV doesn’t cover, it goes in the table.
Video cover letter script. Not a template. A conversation.
Before writing a single word, the tool interviews you. One group of questions at a time. It asks for a concrete result — a number, a measurable outcome, something that didn’t survive the condensing pass. It asks what’s not in the CV at all. It asks what specifically excites you about this company, not the category. It asks how you’d describe your natural communication style.
Only after that does it write anything.
The output is four things: a 60–90 second script (~150–200 words at natural speaking pace), a shorter 45-second version, bullet-point teleprompter cues, and three alternate opening hooks. Because the first line is usually where it collapses — and “Hi, my name is” is the fastest way to lose the next sixty seconds.
A Real Run Against a Stretch Target
A few months ago I picked a posting deliberately — a senior infrastructure role at a well-known hardware company, with positions listed on another continent. A stretch target I had no intention of applying for. I wanted to see how honest the system would be against something it couldn’t rationalize away.
The CV pass was honest. Containers, Kubernetes, CI/CD — solid. The observability tooling they named: I had a functional equivalent, not the exact stack. Customer-facing experience: zero. The hardware acceleration layer — accelerator cluster management, high-performance interconnects — entirely absent from my background.
Five gaps into gaps.md. Three of them hard-required.
Then came the script interview.
First question: one concrete result, with a number. I cited a Kubernetes migration at a bank — fifteen services, zero downtime, two months. Real, verifiable, condensed out of the CV for length.
Second: what’s not in the CV? Two ongoing personal projects — one rebuilding RDMA-capable networking in software, one a structured learning path through GPU programming primitives. Not in the CV because they’re not complete. But real, and they explain why the gap is closing, not static.
The motivation question was where the tool pushed. “Why this company — not the category.” I had to move past “AI infrastructure is the future.” What I ended up writing was closer to: I want to work on the physical substrate that the rest of the stack runs on. That wasn’t in the job description. It was a position I’d been building for months, and the tool forced me to say it out loud clearly.
The script also had to address location. I cited an overseas colleague I’d worked with remotely for eight years — time zones, async handoffs, incidents across continents. “Distributed is already how I operate” isn’t reassurance. It’s a track record.
Final script: 178 words. Specific. In my own language, because the tool had extracted it from me rather than generated it from nothing.
I didn’t apply. The role was the exercise, not the destination. But the gap table that came out of it told me exactly where to point the next three months of learning.
The Gap Map Is the Longer Game
That last part is the one that surprised me.
After three or four applications, gaps.md starts to tell a story. The same skills appear in multiple tables. Hard-required gaps cluster around the same areas. You stop guessing at what to learn next — the market tells you, application by application, precisely where you’re weak.
Most people discard this signal. The CV goes out. The rejection comes back. Nothing gets written down. The same gaps reappear six months later.
gaps.md is just the decision to write it down. Systematically.
The gap table from that stretch application became a learning roadmap. Two items from it are now active projects — not because someone recommended them, but because the process generated receipts.
The Repository
The tailoring command, the gap tracker, and the video script generator are one slash command in Claude Code. The flow is linear: fetch the posting → read the master CV → generate the tailored CV → append gaps → interview the user → write the script.
The repository is here: https://github.com/thenotoriousrunner/tailor-cv/
If you’re already running the master CV workflow from the first post, Steps 7 and 8 are the additions. If you’re starting from scratch, the README covers the full setup.
The CV is the deterministic part. Keyword matching, condensing, reordering — the system handles all of it.
The harder problem is what happens when someone looks at you and asks why.
The tool can’t answer that. But it can ask you the right questions until you can.
VibeOps 
Leave a comment