| name | condense-cv |
| description | Create 1-page or 2-page CV variants from a longer master LaTeX CV. Use when the user asks for a short CV, concise CV, one-page CV, two-page CV, resume-style version, or a role-specific application CV derived from `cv/cv-full.tex` or another master CV source. |
Condense CV
Read the master source first. In this repo, treat cv/cv-full.tex as the master CV and cv/open-source.tex as an auto-generated fragment included by the master (generated from data/open-source.json by scripts/generate_cv_open_source.py). Ignore generated PDFs and LaTeX intermediate files.
Ask for the target page limit and audience if either is missing. Good defaults:
1 page for strict application portals and industry roles
2 pages for research, faculty, and senior technical roles
Prefer creating derived files instead of overwriting the master source. Default output paths:
cv/cv-1page.tex
cv/cv-2page.tex
Add a short header comment to each derived file stating the source file, target page budget, and intended audience.
If both page lengths are requested, derive both from the master source rather than deriving one from the other.
Cut content before tightening formatting. Use this order:
- Remove entire low-priority sections.
- Replace exhaustive lists with selected lists.
- Merge repeated narrative into one-line entries.
- Tighten spacing or typography only if the document is still slightly over budget.
Use references/selection-guide.md for page budgets, pruning order, and audience presets.
Keep these rules:
- Preserve factual accuracy. Do not invent claims, counts, or dates.
- Keep the contact block, current affiliation, and the most relevant research identity.
- Prefer recent, high-signal, role-relevant items over completeness.
- Keep dates, affiliations, named recognitions, and venue names intact.
- Favor selected publications, software, benchmarks, and systems over exhaustive lists.
- Remove teaching, service, student committees, talks, and older internships unless directly relevant.
- The open-source section in cv/open-source.tex is auto-generated from data/open-source.json. For condensed variants, select items inline rather than editing the generated file.
For a 1-page output:
- Keep contact, a 2-4 line summary, current role, education, and 2-4 strongest evidence sections.
- Limit publications to 2-4 items.
- Limit software/projects to 2-4 items.
- Keep awards or funding only if they materially strengthen the application.
- Prefer dense one-line entries and selected lists.
For a 2-page output:
- Keep contact, summary, employment, education, selected awards/funding, selected publications, and selected software/systems.
- Limit publications to 4-8 items.
- Limit software/systems to 3-6 items.
- Add service, teaching, advising, or talks only if the audience values them and space allows.
When the audience is specified, bias selection accordingly:
academia: publications, grants, students, invited talks
industry research / ML: systems, libraries, benchmarks, impact metrics, selected papers
AI security / agent safety: agent-audit, Aegis, TrustLLM, agent security and auditing work
anomaly detection / data mining: PyOD, ADBench, AD-AGENT, anomaly detection papers
If a LaTeX toolchain is available, compile and confirm the final page count. If compilation is unavailable, state that the source was prepared but the page count was not verified locally.
Format Preservation Rule
When an existing derived file (cv-1page.tex or cv-2page.tex) already exists, preserve its format and layout structure. Do not redesign the layout, change the two-column arrangement, alter margin settings, or restructure sections. Only update data within the existing structure:
- Refresh star counts, download counts, and citation numbers from
data/open-source.json and current metrics.
- Swap individual publication or system entries if newer, higher-impact items are available.
- Update award/funding entries with the latest items.
- Shorten entry text only if the update pushes the page count over budget.
The user has approved the existing format. Changing it forces a new round of visual review and approval. Treat the existing layout as a constraint, not a suggestion.