vannifr/agent-skills

Reusable, tool-agnostic skills for coding agents — works the same in Claude
Code, opencode, qwen code, or any tool that reads a SKILL.md bundle. No
tool-specific glue: install a skill by pointing your tool's skill-discovery
path at skills/ /, or symlink it in.
Skills
| Skill | What it does | What makes it different |
|---|---|---|
| nfr-gap-audit | Systematic, self-directed skill-and-NFR audit for a website/web-app project: discovers relevant skills from Tessl and VoltAgent, runs a project-type-aware gap analysis across 12 non-functional-requirement domains (performance, accessibility, security, SEO, testing, reliability, maintainability, scalability, privacy, i18n, monitoring, content quality), and drives phased, regression-checked implementation. Includes a lightweight recheck mode that diffs against a standardized, ID'd baseline. | Combines dual-source skill discovery *and installation* with a 12-domain NFR sweep and a severity/privacy-aware exception path (Critical findings jump the queue; GDPR/legal-risk findings require stakeholder sign-off before any mechanical fix) — no other public skill combines all three. |
| site-review-remediation | End-to-end review *and* remediation of a whole website: full-code review in resumable chunks, data-exposure findings first, fixes as small green commits, every fix validated on live production, CI followed until green. | Where review skills stop at findings and nfr-gap-audit stops at a gap inventory, this one ships and live-validates the fixes, with a pitfalls reference (dropped dotfiles in CI artifacts, unquoted lint globs, proxy caches, old headless browsers) from real runs. |
| site-revision | Full revision of an existing marketing or practitioner website: blind-spot interview and positioning decision, persona walkthrough before and after, content in the customer's own words, a blocking fact-and-quote gate, a live feedback loop with the owner, technical review and a final all-URL check. Ships scripts for the final check, bilingual parity, live verification and image metadata. | Covers what code-focused review skills leave out: positioning, proof and customer language, and a verification gate against invented quotes and anecdotes, with every change validated on production. |
| ci-local-parity | Ensures a project's local verification command (pnpm verify or equivalent) actually covers everything CI checks — adds enforcing git hooks, catches coverage-report gaps, catches CI-image/local binary drift, and keeps trunk-based development honest (checked per push, not per batch). | Solves local↔CI *drift detection* specifically — not "how to build a pipeline" (generic CI-setup guides) or "wire up pre-commit" (dependency-based hook managers). Uses native git hooks, deliberately no Husky/lint-staged dependency. |
| brand-first-impression | Tests what a brand site's first screen makes a visitor understand: blind persona subagents on screenshots, a name-only test, a measured check of where each call to action sits, then a comparison with the written positioning. Report only. | Solves *first-impression and CTA clarity against a stated norm* — narrower than a full persona walkthrough in site-revision, and unlike conversion-audit skills it keeps the personas blind to the norm. |
| ci-hardening | Audits an existing CI pipeline for gates that are green but check nothing: requires a non-zero count in the log and a must-fail proof per gate, mechanical file coverage, strict levels, digest pins and a test-count ratchet. | Solves *gate effectiveness* — complements ci-local-parity (local↔CI drift) and is not a pipeline setup guide. |