Evaluate open-source repos against README promises and render bilingual HTML dossiers.
npx skills add https://github.com/zinan92/repo-evals --skill repo-evals
Please help me install this Agent Skill. Skill: repo-evals Source: https://github.com/zinan92/repo-evals/tree/main Command: npx skills add https://github.com/zinan92/repo-evals --skill repo-evals
repo-evals turns the question “Is this repo/skill actually usable, or just marketing?” into a structured, auditable evaluation dossier with a 0–100 score and an editorial, bilingual product page. It emphasizes claim-first, evidence-backed scoring so readers can challenge specific numbers point-by-point.
Use the repo-evals skill when the user asks to evaluate a repo (e.g., “eval 一下这个 repo”) by providing the target GitHub URL; then follow the framework’s steps to scaffold an evaluation directory, write the claim map, run coverage and scoring, and render the bilingual HTML dossier.
High-intent search queries and answers about installing and using this skill.
To evaluate a repo against its README claims, you convert the stated promises into a claim ledger, then score claim-by-claim outcomes to produce a trustworthy adoption dossier with an auditable 0–100 score.
Claim-first evidence tracking works by mapping README promises to a claim ledger, applying rule-guided static evaluation, and using deterministic verdict calculation to compute a transparent 0–100 score with a six-component breakdown.
Yes, you can compare two repos on consistent dimensions by applying rule-guided static evaluation and evidence provenance tracking, then render the results into a bilingual HTML one-pager covering the decision snapshot and comparison.
You need python3, pyyaml, uuidgen, and gh installed in your environment to run the repo evaluation scripts, track evidence provenance, and generate the bilingual HTML dossier.
The best way to surface claim-versus-reality gaps is to author a claim map from the framework's README, run rule-guided static evaluation against it, and track evidence point-by-point to challenge specific numbers.
You should not use automated repo evaluation when you cannot author a claim map from the README, as the framework requires converting stated promises into a claim ledger before applying rule-guided static evaluation and deterministic verdict calculation.
liutao96Convert GitHub repositories into standardized AI skill scaffolds with SKILL.md and wrapper scripts.
atanakaOrchestrate six sub-skills to derive RM-ODP specifications from business scenarios.
ShubhamsabooGuide structured fact-checking through claim identification, evidence gathering, and credibility assessment.
rahmanayonIdentify missing design decisions in a PLAN and surface concrete additions.
