| name | cv |
| description | CV analysis, Work History Document building, CV tailoring per role, LaTeX CV generation. Use whenever the candidate needs CV work. Read references/fit_scoring.md before tailoring. |
CV Agent
Three modes. Always check prerequisites before any output.
Prerequisites Check
Before generating anything:
Analyse mode:
Tailor mode:
Generate (LaTeX) mode:
If anything is missing: ask for it before writing a single word.
Mode 1 — Analyse
Read the full CV. Output a structured analysis:
## CV Analysis
### What's Strong
- [Credential or bullet that is well-written and impactful]
- [Specific metric or achievement that stands out]
### What's Weak
- [Bullet that is vague — no metric, no action verb, no outcome]
- [Section that is missing or thin]
### What's Missing Entirely
- [Skills expected for [target field] that aren't listed]
- [Certifications or projects that would strengthen the profile]
### Recruiter Perception (honest)
[2-3 sentences: what a recruiter looking at this CV for 10 seconds would think]
### Priority Fixes
1. [Most impactful change]
2. [Second most impactful]
3. [Third]
Mode 2 — Tailor
When tailoring for a specific role:
Step 1 — ATS Keyword Extraction
Extract from the job description:
- Required skills (tools, languages, frameworks)
- Preferred skills
- Action verbs used in the role description
- Degree or GPA requirements
- Any specific domain knowledge mentioned
Step 2 — Gap Analysis
Compare extracted keywords against the candidate's CV and Work History Document.
Output:
## ATS Keyword Gap Report — [Role at Company]
| Keyword | In CV? | Action |
|---|---|---|
| [Tool] | YES | Keep |
| [Framework] | PARTIAL | Strengthen in [role] bullet |
| [Skill] | NO | Add to Skills section if known |
Step 3 — Bullet Rewrites
For each bullet being rewritten, follow this formula:
[Metric or impact] + [action verb] + [what] + [context/scale]
Example:
- Before: "Worked on computer vision pipeline for drone project"
- After: "Built real-time object detection pipeline (YOLOv8, OpenCV) processing 30fps video for autonomous UAV competing in SUAS 2026"
Step 4 — Section Reordering
Move the most relevant experience to the top of each section.
Less relevant entries move down — nothing disappears.
Step 5 — Output
Produce the tailored CV as a markdown file.
Save to data/applications/[company-slug]/cv_tailored.md.
Mode 3 — Generate LaTeX
⚠️ Compiling the .tex file to PDF requires a local LaTeX installation
(TeX Live or MiKTeX). Claude generates the .tex file — you compile it.
If you don't have LaTeX installed, ask for a DOCX version instead.
Load templates/cv_base.tex and populate it with the candidate's data.
ATS-Safe Rules
- Single column — no side-by-side layout
- No tables for main content
- Selectable text only — no rasterized text
- Standard section names: Experience, Education, Projects, Skills
- Black body text — color only for section headings
- Standard fonts: lmodern or helvet
- No icons in the main body
Section Order (default)
- Header (name, contact, links)
- Education
- Experience
- Projects
- Skills
- Certifications & Awards
Adjust order based on the role: if the role is research-focused, put Education
and Projects before Experience.
Output
Save to data/applications/[company-slug]/cv.tex (or data/cv_base.tex for general use).
Print compile command: pdflatex cv.tex (run twice).
Notes on Work History Document
The Work History Document (data/work_history.md) is the source of truth.
It contains the full unedited context, numbers, and proof points behind
every item on the CV — the things that got cut for space.
When tailoring, always pull from the Work History Document first.
If it doesn't exist yet, run /questions to build it before tailoring.