Ding-Ye/learn-codex

learn-codex is a teaching companion that reconstructs the core mechanisms of openai/codex as a sequence of small, incremental Go modules. It's designed for engineers and learners who want to understand why upstream codex-rs makes its architectural choices without reading the full 120+‑crate Rust monorepo. Who uses it: Go developers, systems engineers, and researchers studying agent runtimes and the codex design. Key features: - 10 sequential sessions (s01–s10) that each add one mechanism: minimal event loop, streaming model client, exec tool, V4A patch applier, approval flow, execpolicy DSL, append-only rollout logging with resume, macOS/Linux sandbox adapters, MCP bridge, and CLI driver. - s_full integration walkthrough plus two appendices (mental model, upstream reading map). - Each chapter provides a Go module (code, tests, cmd runner), bilingual docs (EN/CN) with Problem·Solution·HowItWorks sections, and annotated Rust excerpts of the upstream source. - Quick start: Go ≥1.24, run examples and tests. Apache‑2.0 licensed.

このRepoのdescription / GitHub topics / READMEのAI要約に出現したAI関連キーワードをカテゴリ別に表示。各バッジは該当カテゴリの詳細ランキングへリンクします。