Cv Scientist Skills — Security skill for Claude Code
Ten Claude Code skills for image and video ML — data audit, splits, preprocessing, augmentation, model choice, training, post-processing, evaluation — across classification, detection, segmentation, O.
How to install Cv Scientist Skills
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open apoorva2121/cv-scientist-skills and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Cv Scientist Skills does
Ten Claude Code skills for image and video ML — data audit, splits, preprocessing, augmentation, model choice, training, post-processing, evaluation — across classification, detection, segmentation, OCR, medical and satellite imagery. Measured pass rate rose from 0.49 to 0.80.
Alternatives in Security
- Skills — Trail of Bits Claude Code skills for security research, vulnerability detection, and audit workflows 4k ★
- Clawsec — Security skill suite with drift detection, automated audits, and skill integrity verification 798 ★
- Eval Audit — /eval-audit — Batch Skill Evaluation 216 ★
README
Computer Vision Scientist skill set
Ten Claude Code skills that take visual data plus a goal to a validated, deployable model through gated stages: every stage leaves an artefact, screens several methods, carries the best two or three forward, and the final choice comes from a joint search with honest statistics. Built and tested 2026-09-27/28.
The skills
| Stage | Skill | What it decides |
|---|---|---|
| 0, 11 | cv-scientist |
orchestrator: problem framing (incl. provenance, licence, compute), stage gates, time budget, hand-off |
| 1, 3 | cv-data-audit |
decoding/format, label-format conversion, duplicates and leaks, label errors, missing boxes, shortcut cues, shift |
| 2 | cv-validation-design |
grouped / duplicate-aware / multi-label / detection / time splits, test size, holdout |
| 4 | cv-preprocessing |
resolution, crop/letterbox, normalisation, enhancement, serving parity |
| 5 | cv-augmentation |
invariance table, recipes, previews, screening |
| 6 | cv-model-selection |
licence gate, learnability baselines, per-task shortlist (16 task/backbone reference files) |
| 7 | cv-training |
adaptation ladder (probe, LP-FT, LoRA, full fine-tune), recipes, shippable probe, classifier/detector/segmenter templates |
| 8 | cv-postprocessing |
operating points, calibration, conformal sets, NMS, box fusion, TTA, mask cleanup |
| 9 | cv-method-comparison |
paired statistics, proxy screens, joint search (Optuna), ensembles, refit |
| 10 | cv-evaluation |
metrics with intervals, slices, deployment gap, error analysis, explanations, ONNX export and latency |
28 bundled scripts, each tested on real or synthetic data (see `STATUS.md` for every result). Method inventories with verified versions, licences and citations live in each skill's `references/` (sources: `research/R1-R5`, checked 2026-09-27/28).
Install
python tools/install_skills.py --dry-run
python tools/install_skills.pyCopies the ten `cv-*` folders into `~/.claude/skills
Related Skills
Autoconst Claude Bim Health Scorecard
Audit a live Revit model, grade A-F, produce an Excel scorecard. Auto-fixes Uniformat classification. Sourced
Content Audit Skill
Post-publication content maintenance skill for Claude Code — GEO compliance auditing, freshness decay detectio
Agent Audit
Audit a clarc agent file for quality across 8 dimensions — instruction clarity, model appropriateness, tool co
Audit Performance
Performance audit. Detects N+1 queries, memory issues, caching gaps, inefficient loops, batch processing probl
Mad Scientist Skills
Claude Code plugin — 8 skills for architecture, security, observability, optimization, documentation, cognitiv
UX Check
Audit a frontend surface through the Laws of UX and Motion lenses. Fitts's law, Hick's law, Doherty threshold,
Related Agents
Gaia
Remote sensing domain specialist — satellite image processing, spectral analysis, SAR, change detection, time
Inference Integrator
Adapt existing local image ML inference to Electron with faithful preprocessing, process isolation, packaging
Data Builder
Builds and audits datasets for tiny models — synthetic generation, teacher labelling, held-out splits, and lab
