Skill
Data Science CV Repro Reviewer
Review computer-vision experiment reproducibility evidence, dataset readiness, metric gates, and launch risk. Use when a user asks for a cautious CV experime...
When to use Data Science CV Repro Reviewer
Choose if
You want a disciplined reviewer skill that takes a computer-vision experiment package (code version, data version, seed policy, hardware notes, evaluation command, claimed metric) and returns a structured verdict — `reproducible`, `reproducible_with_notes`, `blocked`, or `do_not_promote` — flagging leakage, overfitting, missing baselines, privacy risk, and unsupported claims. Good pre-launch gate before promoting an internal CV result.
Avoid if
You need the skill to actually run the experiment, fetch private data, or operate infrastructure — it's explicitly read-only / instruction-only. Also avoid if the experiment is in a non-CV modality (LLM benchmarks, tabular ML, audio) — the review framework is tuned for computer-vision evidence.
Risk Flags
- LOW scope SKILL.md explicitly states the reviewer must not operate browsers, notebooks, cloud consoles, GPUs, VMs, or storage buckets; it also must not request credentials, tokens, or private dataset access. It is an instruction- only reviewer, not an executor.
- LOW scope SKILL.md flags medical, biometric, face, child-safety, and surveillance-adjacent CV claims as high-risk and demands stronger evidence before issuing a non-blocked verdict.
Cost
Type: Free
Distribution
- ClawHub
data-science-cv-repro-lab- License
- MIT-0