Loading hundreds of long agent instructions at once wastes context and causes silent truncation. I wanted a verified catalogue where you install only the category or skill your task actually needs.
Load only the instructions the task needs.
Markdown defines each skill in its dedicated category repository. Python validates frontmatter, builds SHA-256 verified bundles, and compiles the SQLite FTS5 and JSON indices.
The catalogue is public open source. Installing a skill downloads Markdown files from GitHub; executing a skill uses your host agent's local tools and permissions.
Skills are structured instructions, not runtime sandboxes. Always review a skill's instructions and your agent's tool permissions before execution.
Dumping dozens of massive system prompts into a coding agent wastes context window tokens and makes the model forget instructions halfway through a task.
Instead of loading everything at once, I wanted a clean library where an engineer or agent pulls only the exact playbook needed for the job at hand.
Skillary organizes 277 verified skills across 12 focused category repositories so your agent loads concise, task-specific instructions on demand.
Search by task in the browser, read the plain Markdown instructions before trusting them, and copy a one-line install command for Claude Code, Gemini, Antigravity, Codex, or Cursor.