Proving sdocs.dev is running the public code
SHA-256 hash verification via CI, but reproducible builds already solve this.
Deploy and Self Heal any GitHub repo
It uses an LLM not just for scaffolding but as an active repair agent: Claude generates Dockerfiles, diagnoses crashes (diagnoseCrash) and can trigger redeploys or memory scaling. The architecture ties BullMQ, Fly Machines and a 30s health loop into a real feedback loop — clever and ambitious, but it also leans heavily on Claude + Fly/Supabase so expect nontrivial setup and vendor/LLM dependency.
Backend engineers, DevOps/Platform engineers, indie teams who deploy apps and want automated ops
SHA-256 hash verification via CI, but reproducible builds already solve this.
MCP integration exposes infrastructure as AI tools — clever, but Coolify exists.
Feels like Codespaces for LLM-driven coding: each session gets its own git worktree plus a browser xterm and a virtual mobile keyboard. The background 'shepherd' that preserves PTY sessions across server restarts and the single Go binary embedding the React frontend are neat, practical engineering decisions — just remember you need the Claude/Codex CLIs and to lock down access when exposing the server.
Spawns dozens of Claude Code agents in tmux with auto-recovery and shared memory—neat hack, niche audience.
GitHub sync for Claude Code configs with secret scanning before commit.
TDD-first auto-fix approach writes failing tests before generating patches, not just patching blindly.