KernelLord/pickysteve

Local-first skill router and context picker for coding agents — hybrid BM25 + embedding retrieval (RRF-fused), cross-encoder rerank with a calibrated floor, and an LLM judge select the right skill, then assemble a minimal, nonce-boundaried context bundle for the execution model. Built-in ONNX prompt-injection gate scans both the incoming request and every retrieved document, fail-closed (100% detection / 0 bypasses on a 180-payload red-team corpus). Runs entirely on local Ollama by default (any OpenAI-compatible endpoint also works). Ships as an MCP stdio server or an OpenAI-compatible proxy; one-command installer wires up 18 coding agents including Claude Code, Codex, Cursor, and Windsurf. Install: `git clone` + `uv` (PyPI package not yet published). MIT.

Project

Browse this MCP server in the TensorBlock MCP Index.