Give every claim a source
Start with your master resume and details you've confirmed. Mark missing information for review instead of inventing a metric. Check a public pull request's merge state before describing it as merged.
A clear CV built from your real work, in your own words.
MIT licensed. Works with your existing agent.
Actual one-page A4 output from the skill, using fictional sample data.
Scroll inside the preview to read the page. Fit page shows the full layout; Open PDF provides selectable text and your browser's zoom controls.
Choose relevant work without changing the facts.
Start with your master resume and details you've confirmed. Mark missing information for review instead of inventing a metric. Check a public pull request's merge state before describing it as merged.
Your selected original writing can guide phrasing and rhythm through a private voice reference. Resume facts stay separate. The overlap script helps review shared wording; accurate technical terms do not need disguising.
Keep the latest approved CV as the design baseline. Check extracted text and section order, compare the exported pages with that reference, and review employer autofill when offered. Parser behavior varies, so these checks cannot certify every ATS.
Your foundation.md resume and provenance.md verification notes are excluded from version control. Your agent provider's privacy policy still applies when it reads them.
The script flags shared runs of four or more words in a CV and a posting, reporting repeated matches at each location. Review each match in context: common technical terms are not proof of copying.
This skills label can stay if it accurately describes your work. Rephrasing it is optional.
Skills label in both the CV and the posting
CLI tools and libraries
Shared run found: [4 words] cli tools and libraries
Review the wording in context. An overlap isn't proof of copying.
A shorter alternative for the same skills label
Command-line tooling
No shared run of 4+ words in this short example.
The experience stays the same. Only the label changes.
This switches between sample text. Run the Python script with your own files to check a draft.
You review the facts and wording before the agent exports a file.
Share the posting, verified sources and latest approved CV. The existing CV is the default design baseline; the agent asks before changing the look or section order.
Approve wording and any proposed design improvements separately. Nothing about the visual structure changes without your approval.
Verify required text and section landmarks in extracted PDF text, compare each page with the approved CV, and review employer autofill when available. These checks do not certify every ATS.
Explore the animated workflow, source map and fictional career-lane PDFs.
Add the skill to your agent, then review your first draft.
Local exports are tested on Linux and macOS with Python 3.11+, a Chromium-family browser, and Poppler's pdftotext and pdfinfo. GitHub CLI is used when checking public contributions.
Rezi can be an optional resume source. Amazon Quick is an optional DOCX workflow; the skill does not bundle either service.
git clone https://github.com/chindris-mihai-alexandru/cv-witness.git ~/.adal/skills/cv-witness
Run in your terminal. Private resume files are not included in the clone.
/plugin marketplace add chindris-mihai-alexandru/cv-witness
/plugin install cv-witness@cv-witness
Send these as two separate commands inside Claude Code, not in your shell.
git clone https://github.com/chindris-mihai-alexandru/cv-witness.git ~/.agents/skills/cv-witness
Run in your terminal. Codex discovers personal skills in this directory.
git clone https://github.com/chindris-mihai-alexandru/cv-witness.git
Ask your agent to read the root SKILL.md. Discovery paths depend on the host.
Already have a clone at that path? Update it rather than overwriting it. See the setup notes for requirements and private sources.
No. Parsers and employer configurations differ. The text-based layout and export checks help catch formatting problems, but they cannot guarantee correct parsing, a ranking or an interview.
No. This is a product overview with a rendered fictional sample, not a resume editor or upload form. The skill runs through your chosen agent. That provider's privacy and data policies apply to any sources you share with it.
You can now build a private style reference from selected original Markdown, Cabinet pages, Obsidian notes, Sanity exports, or normalized original-text records. It describes writing habits rather than training a model. Generated memories and assistant prose are excluded when labelled; you still confirm authorship and review the draft. Exact voice matching is not promised. Read the voice corpus guide.
Engineering, Operations/Support, Research and Management presets reorder complete sections in one single-column layout. Roles repeat without fixed slots, and A4 or Letter is explicit. Longer CVs need an approved page limit; no content is dropped to force a pass. Inspect the fictional template gallery.
The skill is MIT licensed and does not require Rezi or Amazon Quick. You still need an agent host, and any charges or requirements of that host are separate.