juliusbrussee/caveman-code
7 skills142 installs2.7k starsGitHub
Install
npx skills add https://github.com/juliusbrussee/caveman-codeSkills in this repo
1Caveman CompressCaveman Compress is an agent skill that turns on ultra-compressed communication so solo builders burn far fewer tokens per turn without losing accuracy. When someone says caveman mode, less tokens, be brief, or the context window is nearly full, the assistant drops articles and filler, uses short synonyms, and answers in tight fragments while leaving code, APIs, and error strings verbatim. Full mode is the default; lite and ultra dial aggression up or down. It fits indie developers running long Claude Code, Cursor, or Codex sessions on big repos where every reply competes with file reads and tool output for the same budget. The skill does not run shell or network—it only changes prose style around unchanged code blocks. Use it journey-wide whenever chat cost matters; skip it when you need polished copy for customers, investors, or published docs.24installs2Cavekit Design SystemCavekit Design System is an agent skill for writing and maintaining DESIGN.md as the visual specification layer for Cavekit projects. Solo and indie builders use it when they want one authoritative document that coding agents read before building screens, instead of fighting inconsistent styling across sessions. The skill teaches a nine-section format covering design tokens, component appearance, responsive behavior, and accessibility, positioned next to behavioral kits and build plans. It fits defining a new visual identity, folding in an external design system, or reviewing existing UI for token violations. Because DESIGN.md is a living contract, the skill emphasizes centralizing visual decisions so every UI-focused agent applies the same language. Medium effort; read, grep, and edit tooling only.20installs3Cavekit MethodologyCavekit Methodology is a journey-wide agent skill for specification-driven development with AI coding agents. It encodes the Hunt lifecycle from Draft through Monitor and insists that teams define what they want—usually as kits—before telling agents how to implement. Solo builders use it when starting a Cavekit project, restructuring a legacy codebase into kits and plans, or deciding which sub-skill to invoke next. Kits act as a living contract: structured, auditable, and evolvable without locking you to one framework. The methodology applies equally to greenfield products and rewrites because both paths flow through kits to generated code. Read and grep tooling reflect guidance-heavy routing rather than automated refactors. Use it whenever you catch yourself pasting requirements straight into a build prompt.20installs4Cavekit RevisionCavekit Revision is an agent skill for solo builders who run Cavekit-style iteration loops and keep hitting the same bugs after manual fixes. Instead of patching application code in isolation, you trace each failure upstream through kits and prompts until you find the specification gap that allowed it, then fix at the source so the next loop reproduces the correction autonomously. The skill packages a six-step revision process plus a single-failure backpropagation protocol for when convergence stalls, a hot-fix landed without a matching kit change, or a bug class keeps resurfacing. It is procedural knowledge for self-correcting agent workflows: kits become progressively complete and the distance between documented intent and working software shrinks over iterations. Use it after review or testing exposes a defect, during build when the loop diverges from plans, or in operate when production issues map back to weak prompts. It does not replace root-cause debugging in runtime logs—it complements it by ensuring fixes persist in the artifact chain your agent actually follows.20installs5Cavekit Validation FirstCavekit Validation-First is an agent skill for solo and indie builders using spec-driven development with Cavekit. It encodes a validation-first design doctrine: every kit requirement must ship with acceptance criteria that an agent—or CI—can verify automatically, because non-deterministic agents otherwise produce output nobody can prove correct. The skill documents a six-gate validation sequence ordered from fast, cheap failures to more expensive checks, plus phase gates, merge protocol, completion signals, and reusable acceptance-criteria patterns. Use it when you are writing acceptance criteria for a new feature, wiring CI gates to a plan, or debugging why an agent keeps marking work done without objective proof. It pairs read-only tooling (read, grep) with procedural knowledge rather than executing builds itself, so you still wire gates in your repo—but the skill tells you what must exist at each layer so SDD does not collapse into narrative specs.20installs6Dynamic ResourcesDynamic-resources is an agent skill in the caveman-code repo for defining and shipping a coding-agent color theme as structured JSON. Solo and indie builders who run pi-mono style interactive coding agents use it when the default terminal palette feels flat or when tool-call states are hard to scan at a glance. The skill centers on a single theme document with vars for cyan, blue, green, red, yellow, grays, and role-specific backgrounds for user messages, custom messages, and tool outcomes. Colors map those vars into borders, accents, warnings, markdown elements, and diff styling so agent transcripts stay readable during long sessions. Install it when you are polishing the agent shell alongside your project, not when you need marketing pages or production observability. Confidence is moderate because public docs emphasize the theme payload rather than a full procedural SKILL.md checklist.19installs7Plugin CreatorPlugin Creator is an agent skill for solo and indie builders who ship extensions in the Cave ecosystem and want a correct bundle on the first try. Instead of hand-rolling folders and guessing manifest fields, it scaffolds a complete plugin tree anchored by `.cave-plugin/plugin.json`, with optional `commands/`, `skills/`, `agents/`, `themes/`, and `hooks/` directories sized to what you actually enable in `capabilities`. The embedded schema covers semver versioning, kebab-case naming, `caveVersion` compatibility (for example `>=0.65.0`), licensing, homepage URLs, and free-form tags that power `caveman plugin search`. Use it when you are starting a new marketplace plugin, packaging commands or skills for reuse, or publishing from GitHub or a zip URL. It keeps capability flags honest so installers know whether commands, skills, agents, themes, MCP, or hooks are included—reducing broken installs and review churn before you publish.19installs