adapt-to-stacklisted

# Adapt to stack — one specialist per layer, generated from one file A repo declares its layer chain once, in its own `CLAUDE.md`. This turns that declaration into the agent files and skill files the pipeline dispatches to. > **This is not a stage. It is one step**, and it runs again — unchanged — every time the > chain gains a layer. The first run happens on a repo that has just been scaffolded. The > second may be months later. **Choose nothing.** Every fact you write into a generated file is already written down one file away. If you find yourself deciding what a layer is, where it lives, what it builds with, or which model it should run on, you are reading the wrong file. **Stop and say so.** ## The one input **Read the repo's `CLAUDE.md`. That is the only input.** No arguments, no prompts, nothing asked of the human. A prompt would state the stack a second time, and two copies can disagree. | You need | Section of `CLAUDE.md` | |---|---| | The layers, in order, and where each one lives | the chain section | | Each layer's build / test commands | *build / test / run* | | Each layer's upstream and downstream contract | the chain **order** — the neighbours either side | | A model per layer, **if one is stated** | the chain section's optional model lines | ### Which `CLAUDE.md` **The one in the repo you are standing in, and never the workspace file.** Run inside a repo, you generate that repo's layers from that repo's own file. `repo.CLAUDE.md` ships two chain shapes