Ding-Ye/learn-mini-swe-agent

A compact, chaptered Go re‑implementation of SWE-agent/mini-swe-agent that teaches how the "100‑line AI coding agent" works by rebuilding it piece‑by‑piece. Intended for developers and learners who want a hands‑on, minimal mental model of agent internals. What it is for - Learn each upstream mechanism by writing a small Go reimplementation and reading annotated upstream excerpts. Who uses it - Engineers, students, and researchers studying agent architecture and SWE-bench style agents. Key features - 8 sessions (s01–s08) + appendices; each session is a standalone Go module (Go 1.23+), runnable and tested. - Pedagogy: mental model → 30–100 line code excerpt → diff → side‑by‑side upstream source reading. - Implements: minimal agent loop, Provider interface, action parsing, templating/config merging, control flow signals, Docker sandbox, human-in-loop modes, provider composition. - Docs, tests, and annotated upstream readings. MIT licensed.