AI Agent Skills

Four. humanizar and human-ai each ask 8 Score and 12 Noul questions, slop-eval asks 8 Score and 16 Noul, and agent-plugin-eval asks 7 Score and 19 Noul. The remaining skills are ordinary prompt-and-reference skills with no model call at all.

Nothing breaks. Each Jev-enabled skill ships a discovery protocol that checks jev_available() before use and falls back to heuristic methods, so the skill still works with the key absent. Live calls need a key, and the integration is optional rather than required.

Subjective quality axes rather than facts: pattern severity, voice match quality, rewrite confidence, tell severity, section classification, UX coherence, documentation clarity, schema design, and secret detection. Each skill carries its own question definitions in scripts/jev_questions.json plus a reference guide with integration patterns and code examples.

Documented installs are repository-level: skillkit install fabricioctelles/skills, or npx agent-skills-cli install with the repo URL, and an agent flag to target specific editors such as claude-code and cursor. Whether a single skill can be installed on its own is not stated in the source.

Agent Skills are a lightweight open format: each skill is a folder containing a SKILL.md with metadata and instructions that an agent loads on demand, so nothing is resident in your context until it is needed. The collection targets Claude Code, Cursor, Kiro, Windsurf and OpenCode, and installation is one command, with an --agent flag when you want to target specific editors. There is also a separate OpenAI plugin, Deslop Writing, bundling the English and Brazilian Portuguese editing workflows for ChatGPT and Codex.

Twenty-two more, spanning engineering, SEO, GEO and infrastructure: a startup idea analyser that runs a Paul Graham, Dan Koe and Seth Godin lens on an idea, a generative engine optimisation skill for getting cited in AI answers, a Substack platform expert covering formatting, SEO, Notes and conversion, a Pier Cloud API guide, an ultimate design system master with ten role-play prompts, and a self-contained port of Lauren Tan's pstack with 23 playbooks. Most carry a Security A badge from Skills Directory.

Apache-2.0 across the collection except where a skill carries its own license. The one restriction worth knowing is skills/revenue-centric-design, which is source-available rather than open and prohibits gambling, betting or casino use. The Jev integration itself is optional throughout: live calls need a key, and each skill checks jev_available() and falls back to heuristic methods when there is not one.