Limpiador is a command line agent that takes on the repository upkeep engineers keep postponing. It reasons about a codebase the way a careful reviewer would, then acts through 56 tools across five namespaces covering git, GitHub, the filesystem, AST analysis with tree-sitter, and test and CI workflows. Tools load dynamically from a searchable registry, so the model only ever sees what the current task needs, which is what keeps it usable on large repositories instead of drowning in context. A separate reviewer subagent runs against a scoped read-only registry and checks the work before anything lands, and context compaction with explicit eviction lets long sessions keep going without losing the thread. A four-tier test pyramid covers the whole system. Built in Python against the OpenAI API, and still actively maintained. It is the project I point to when someone asks whether an agent can do real engineering work rather than just answer questions.