Expert playbooks for your agents,one arrow each
Search the skills.sh directory in the app, read a skill before you install it, and connect it to the agents that should follow it. They load it by themselves when a task matches.
20,000
most popular skills from skills.sh, searchable offline
4
security scanners’ results shown before you install
0
skills an agent sees without an arrow to it
The agent notices the skill on its own
A connected skill appears in the agent’s tool list with its name and when to use it. When your request matches, the agent loads it first and follows it — you do not have to mention the skill.
brainstorming
obra/superpowers
You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.
Connected agents: 0
Draw an arrow from this card to an agent.
The skills people actually install
Popular skills first, with their install counts. Open one to read its SKILL.md, the list of files, and security scans from four scanners — before anything lands on your disk.
Try it: type “design” or “debug”, open a skill, install it.
Pick one to see what it is and exactly what will run.
Read it, install it, connect it
Skills are folders with a SKILL.md: instructions, sometimes scripts and references. NeuroSquad keeps them in its own folder, never in your ~/.claude.
Search skills.sh
Instant local search over the most popular skills, merged with skills.sh’s own search and install counts when you are online.
Read before installing
The whole SKILL.md, the files it ships and the security scans. Skills with scripts are flagged — an agent following them may run them.
Pinned on install
The files you read are exactly the files installed, with a checksum. An update is a new version you review again.
Arrow it to an agent
The agent sees the skill’s name and “when to use it” in skill_load, loads the full text when a task matches, and reads its files with skill_read_file.
Same skills for every supported harness
Skills reach agents through NeuroSquad’s own MCP connection — not through each harness’s skills folder — so one install works for all of them.
- Claude CodeRight awayThe skill list updates mid-session; checked with a real agent that loaded a skill unprompted.
- Codex CLIOn its next startPicks up changes when it starts; restart the card after drawing the arrow.
- Qwen CodeOn its next startSame as Codex: after the next start.
- OpenCode, Cursor, Kilo, Hermes, Pi, CrushNot supportedNot connected to the app’s MCP server; the card says so.
Skill text goes straight into a model — so you read it first
A skill is instructions an agent will follow. Everything from the catalog is treated as untrusted.
Review before install
The full SKILL.md, every file and its size, and the scanners’ verdicts. Nothing is written until you confirm.
Scripts are flagged
Skills that ship shell, Python or JavaScript files are marked, so you know an agent may run them.
Safe unpacking
Every path is checked — no .., no absolute paths — and sizes are capped. The folder is written aside and moved into place.
Only where you point it
An agent without an arrow to the skill card sees nothing of it. Delete the arrow and it is gone from that agent.
Questions
Teach your agents how your team works
Install NeuroSquad, open the catalog, arrow a skill to an agent.