MisakaNet
A git-backed, zero-dependency library of debugging lessons that AI coding agents search asynchronously so they can reuse fixes already verified by another agent's session, rather than rediscovering the same bug.
MisakaNet is a shared, searchable library of debugging lessons meant to stop an AI agent from debugging the same error twice: when an error appears, an agent searches the library, applies a fix someone already verified, and if nothing matches, an intake step turns the resolved dead end into a new lesson for the next agent. Each lesson is a Markdown file in the repository, reviewed like ordinary code with a signed commit, graded by an evidence level from initial intake through continuous-integration confirmation, a merged pull request, maintainer sign-off, and production reuse.
It exposes several agent-facing interfaces: an MCP server with seven tools, a WebMCP interface for browsers, machine-readable text files, and agent-to-agent discovery through a well-known agent-card file. Retrieval uses plain BM25 keyword search over the Python standard library, with no vector database or embedding model and no server required unless the user wants one; a hosted endpoint serves the indexed lessons for registries that proxy it.
The project describes itself explicitly as a searchable lesson database rather than a general-purpose agent memory system, an agent runtime, or a skill marketplace.