Meta Skill
A local-first skill management platform for AI coding agents, combining SQLite and Git persistence, hybrid semantic search, bandit-optimized suggestions and MCP integration to turn operational knowledge into reusable, searchable artifacts.
Meta Skill (ms) stores skills with dual persistence, SQLite for queries and Git for an audit trail, and finds them again with hybrid BM25 and hash-embedding search fused by reciprocal rank fusion. A UCB bandit algorithm adapts its suggestions from feedback over time, and dependency graph analysis surfaces cycles, bottlenecks and execution plans across skills using PageRank-style ranking.
It also includes prompt-injection defense and destructive-command safety gates, provenance tracking that links a skill back to the session it came from, Git-based multi-machine synchronization with conflict resolution, portable bundle distribution with checksums, automatic mining of anti-patterns from session history, and token-packing to fit a set of skills within a context budget. Skills can come from hand-written SKILL.md files, mined sessions, imported bundles or guided workflows, and an MCP server exposes them as native tools for Claude, Codex and other agents. Rust, requires a pinned nightly toolchain.