← ClaudeAtlas

implementing-submoduleslisted

Implement a documented-but-empty Core or AWS submodule of @m3l-automation/m3l-common end-to-end from its docs/reference spec, under strict TDD + hub-and-spoke. Use for "implement"/"build"/"flesh out" a submodule that already has a docs/reference/{core,aws}/<name>.md page, even named casually ("the retry stuff"). No spec page yet → scaffolding-submodules first. Docs stance: context7 MCP (ADR-0093).
monte3l/m3l-automation · ★ 0 · AI & Automation · score 56
Install: claude install-skill monte3l/m3l-automation
# implementing-submodules This skill is the **hub playbook** for turning a `docs/reference` page into real, reviewed, tested library code. The library is a fully-documented but empty scaffold; each `docs/reference/{core,aws}/<module>.md` is the authoritative contract for what its submodule must export and how it must behave. ## Operating model: you are the hub, not a worker You (the main agent) **coordinate only**. You do **not** write `src/` code or test code yourself, and you do **not** review code. Every substantive step runs in an **isolated spoke subagent** with the right tool grants. This makes "the writer is never the reviewer" a structural guarantee, not a polite request, and keeps your context lean across what is often a long loop. Spokes don't share memory with each other or persist between sessions, so two things matter: pass each spoke **explicit context** (target module, the contract, concrete file paths), and record progress in the durable state file `docs/implementation-status.md` after every phase. | Phase | Spoke (subagent) | Writes | Hand it | | --------------- | -----------------------------------------------------------------------------------------------------------------------------------