⚡ hyperskills
Focused AI agent skills for things models don't already know
Knowledge, guidance, wisdom, SOTA. Reach for what fits.
💎 What This Is
Models already know how to write React components, Kubernetes manifests, and PyTorch code. They don't need 300 lines of examples for that.
hyperskills is built around an agent workflow. Brainstorming structured by the Double Diamond. Wave-based research with deferred synthesis. Verification-driven planning and implementation. Cross-model peer review that catches what self-review misses. Six orchestration strategies for multi-agent work. Conversation consolidation that pulls signal out of past sessions into persistent memory. The process skills are the heart of it, mined from thousands of real dispatches and tens of thousands of tracked operations.
Domain skills round out the toolbox where models have stale or missing knowledge: the agent-sandbox Kubernetes operator (too new for training data), current Astral Python tooling, Tilt operational decision trees, terminal UI design that survives across emulators, an OWASP/SLSA security reference.
Each skill encodes procedural knowledge, decision trees, anti-patterns, and current SOTA. None prescribes a strict workflow. They give you knowledge and framing; you decide when to reach for them. Skills carry procedural knowledge in-context; Sibyl carries decisions, patterns, and learnings across sessions. 17 skills, all installable independently.
⚡ Installation
Claude Code
# Register the marketplace, then install
/plugin marketplace add hyperb1iss/hyperskills
/plugin install hyperskills@hyperb1iss
Vercel Skills (skills.sh)
# All skills
npx skills add hyperbliss/hyperskills --all
# Pick what you need
npx skills add hyperbliss/hyperskills --skill implement
npx skills add hyperbliss/hyperskills --skill orchestrate
Manual
git clone https://github.com/hyperb1iss/hyperskills.git
ln -s $(pwd)/hyperskills/skills ~/.claude/skills/hyperskills
🪄 Composing Skills
Skills are independent. None of them require the others. A typo fix doesn't need brainstorming, a clear bug doesn't need research, and the Python tooling skills compose freely.
A few combinations come up often, more as observation than prescription:
| Situation | Skills that pair well |
|---|---|
| New feature | brainstorm, plan, implement, cross-model-review |
| Greenfield project | brainstorm, research, plan, orchestrate, implement |
| Architecture decision | brainstorm, research |
| Large refactor | plan, orchestrate, implement, cross-model-review |
| Bug fix | implement (the skill scales itself for trivial fixes) |
| Python project work | uv, ruff, ty, uv-build |
| Knowledge consolidation | dream pulls insights from past sessions into Sibyl |
Domain skills (security, git, tilt, agent-sandbox, tui-design, uv, ruff, ty, uv-build) plug in wherever the work touches their territory. Any skill can loop back when new questions emerge.
🔮 Skills
Process Skills
How to approach a class of work: workflows, phases, decision gates. The interesting part of hyperskills lives here.
brainstorm: Structured Ideation
Double Diamond model for creative work. Diverge on the problem, converge on a definition, diverge on solutions, converge on a decision. Grounded in Sibyl so you don't re-explore solved problems. Includes a Council pattern (advocate / critic agents) for complex architectural decisions.
/hyperskills:brainstorm