Show HN: Elliot – my coding agent that helps me code, and sometimes helps itself (github.com)

Elliot is a new Python package and CLI that orchestrates a team of specialist coding agents (built on openai-agents) to help developers plan, inspect, edit, and maintain codebases. It runs locally on Python 3.12+, only requires network access to an OpenAI-compatible endpoint you configure, and produces rich Markdown logs in the terminal via rich. Elliot maintains a live task plan, spawns scoped helper sub-agents, and presents updates inline; it also implements tooling safeguards that prompt for confirmation before file writes or executing external commands. You can run it from a temporary environment (uvx), install via pip/pipx, or invoke python -m elliot, and control orchestration length with --max-turns. Set OPENAI_API_KEY (required), optional OPENAI_BASE_URL and OPENAI_DEFAULT_MODEL (defaults togpt-5). Technically, Elliot provides a curated toolbelt for sub-agents: structural code search and rewrites via ast-grep (code_search, code_rewrite), directory/file inspection (list_dir, file_tail, read_slice), previewable sed-based edits, git_command with confirmation, python_run for executing snippets, Ruff lint/format, ask_user for clarification, and ask_expert that uses the OpenAI Responses API for concise guidance. All tool interactions are logged in Markdown, core logic lives under elliot/ (agent, plan, tools, cli, output), and the project is MIT-licensed. For AI/ML engineers this offers a composable, auditable local orchestration layer for model-driven code work with built-in safety and extensibility.