paulrberg/agent-skills
11 skills18k installs682 starsGitHub
Install
npx skills add https://github.com/paulrberg/agent-skillsSkills in this repo
1Code SimplifyCode Simplify is an agent skill for solo builders who want messy or clever code rewritten into maintainable, explicit forms without changing what callers observe. It runs inside a git repo, resolves targets from your instructions or from files changed in the current session, and otherwise falls back to uncommitted tracked and untracked paths. The agent simplifies while honoring public contracts and side effects, then—unless you pass --no-verify or --no-report—checks that behavior still holds and summarizes what changed. That makes it useful right after a feature spike in Build, or as a structured cleanup pass in Ship before review or release. It pairs well with agents that tend toward compressed one-liners or nested logic. It is not a substitute for product-level architecture decisions or for security audits; it optimizes local clarity and long-term readability on a defined slice of the tree.2.1kinstalls2Effect Tseffect-ts is an agent skill focused on configuration and environment variables in TypeScript using the Effect library’s Config module. Solo builders shipping APIs, CLIs, or agent backends install it when agents otherwise read process.env directly, leak API keys in logs, or scatter DATABASE_* parsing. The skill shows required vs defaulted strings and numbers, Redacted for sensitive keys, nested prefix grouping, and Effect.gen programs that yield typed config. It also covers injecting ConfigProvider.fromMap for tests and local overrides via Layer.setConfigProvider—patterns that matter from first backend scaffold through production operate iterations. Complexity is advanced because Effect’s functional effect model assumes familiarity with generators and layers. Use whenever SKILL guidance says to pull config handling into Effect, not ad-hoc dotenv wrappers.1.8kinstalls3Code ReviewThis code-review skill packages focused review profiles for infrastructure configuration and structured data code paths. Solo builders invoke it when a PR touches timeouts, pools, env-specific values, feature flags, observability, or parsers for CSV, YAML, JSON, and binary formats. The configuration profile flags high-magnitude changes without justification, inverted timeout/retry hierarchies, pool/limit starvation, blind prod copies from dev, missing rollback paths, and observability gaps. The data formats profile targets formula injection in CSV exports, unsafe YAML loaders, schema-free parsing, numeric precision loss, unvalidated binary parsers, and encoding ambiguity. Reviewers are steered to cite rule IDs and expected evidence such as before/after values and load failure scenarios. It complements generic chat review with enumerated, severity-ranked checks suitable for agent-assisted PR triage before ship.1.8kinstalls4Cli Ghcli-gh packages a bash-driven GitHub pull request workflow for solo and indie builders who ship with git and the GitHub CLI. The included auto-pr-create script walks you through branch naming tied to tickets, pattern-based label assignment, optional draft PRs, and safe branch checkout before opening the PR against a configurable default base (main). It is aimed at agents or developers who want repeatable PR hygiene without clicking through the GitHub UI. Use it when you have commits ready on a feature branch and need a consistent PR title, labels, and gh pr create invocation. The skill fits small teams and single-maintainer repos where process still matters but tooling should stay lightweight. Pair it with your existing commit and CI habits; review the script paths and gh auth on your machine before running in production repos.1.7kinstalls5Code Polishcode-polish is a workflow skill for solo builders who want one command instead of chaining simplify and review manually on messy recent edits. It verifies a git repository, resolves scope to user-provided paths, a commit range, a Resolved scope block, or session-modified files—with fallback to all uncommitted tracked and untracked changes when the session list is empty. It then runs code-simplify for readability and maintainability, followed by code-review --fix for correctness, security, and quality with automatic fixes. The design optimizes for a single scope resolution, one report, and no duplicate simplify passes. Use it when you ask to polish, simplify and review, refactor and review, or clean up and fix code you have just touched. It sits in Ship review but also helps late Build cleanup before you open a PR.1.6kinstalls6Tailwind CssTailwind CSS is an agent skill that encodes paulrberg-style coding preferences for Tailwind utility workflows. Solo and indie builders shipping SaaS dashboards, marketing sites, or browser extensions install it so the agent applies opinionated layout, spacing, typography, color, border, gradient, arbitrary value, class-merging, image-sizing, z-index, and dark-mode rules instead of generic Tailwind guesses. The guidance is procedural: use gap on flex and grid rather than space utilities that fail on wrap, prefer size-* for equal dimensions, always use min-h-dvh over min-h-screen for real mobile viewports, favor predictable margin direction, and pad parents instead of bottom-margin hacks on last children. It fits whenever you are authoring or debugging class strings in React or similar stacks during Build, and it still helps during Ship review passes when you are cleaning up inconsistent utilities before release.1.6kinstalls7Md Docsmd-docs is an agent skill for solo maintainers of open-source libraries, CLIs, and agent skill repos who keep confusing humans and models with one overloaded README. It enforces a strict audience split: README.md markets the project—description, badges, doc links, license, contributing pointer—while AGENTS.md (often symlinked from CLAUDE.md) holds the operational truth agents need: stack, package scripts, just recipes, test and lint flows, and repo-specific rules. Every documentation workflow in the pack must respect that boundary so npm/GitHub visitors are not buried in agent jargon, and coding agents are not starved of commands. Use it during initial repo setup, before release polish, and when onboarding Cursor or Claude Code to a mature codebase.1.6kinstalls8Bump ReleaseBump Release is an agent skill that encodes the stripped Common Changelog spec so solo maintainers produce CHANGELOG.md releases that read well for humans, not just machines. It enforces latest-first semver ordering, mandatory entries for stable releases, and second-level release headers with ISO dates matching git tags (optional v prefix on tags only). Writers sort by user impact, omit trivial churn, and attach reference links—typically GitHub release URLs—for traceability. Notices handle breaking upgrades or inaugural 1.0.0 moments without burying them in bullet lists. The skill fits indie open-source and small product repos where the founder tags v1.0.1 and needs consistent Fixed/Added language the same day. Pair it with your existing bump/version script mentally: this skill governs narrative structure, not necessarily automating git tag pushes unless your agent workflow does that separately. Use when preparing a public release story; skip for internal-only deploy logs with no semver policy.1.6kinstalls9Bump Depsbump-deps wraps the taze CLI so a solo builder can audit dependency versions without sitting through interactive prompts. The bundled run-taze script checks that taze is available, verifies package.json in the chosen directory, detects monorepos via workspaces or pnpm-workspace.yaml, and runs taze major with locked versions included so major bumps are visible before you merge. Optional include lists narrow scope; the skill is oriented toward high-effort, deliberate upgrade sessions with dry-run style hints in its invocation contract. It fits indie maintainers who ship frequently and need a repeatable agent-driven bump pass across single packages or entire workspaces. It does not replace reading changelogs, running your test suite after upgrades, or pinning policy decisions—that remains on you after taze reports candidates.1.5kinstalls10YeetYeet is an agent skill for posting comments on existing GitHub issues or pull requests. On GitHub’s model, pull requests are issues with extras, so one `gh issue comment` path covers both. You pass flexible identifiers—a full URL, `owner/repo#number`, or a number inferred from the current repo—plus free-text comment context describing what you want to say. The workflow validates prerequisites, resolves the target issue, reads thread context, then composes a reply in an informal voice aligned with the repo’s create-issue skill. Solo and indie builders use it to unblock PR review threads, acknowledge bugs, or drop quick updates without leaving the agent session to craft gh invocations by hand.1.5kinstalls11Biome Jsbiome-js is an agent skill that teaches how to set up and extend BiomeJS the way Sablier-style projects do: root biome.jsonc with schema path, extends from shared devkit packages, and always-explicit files.includes globs. Solo builders shipping TypeScript libraries or React UI monorepos use it when they need one formatter-linter instead of ESLint plus Prettier, or when they must inherit root config in packages/my-package with // and package-specific overrides. The skill covers extending base versus base+ui configs, common include patterns per project type, and operational fixes when Biome errors block CI or lint-staged. It fits indie teams who want fast, opinionated lint/format defaults without maintaining parallel toolchains.1.1kinstalls