← ClaudeAtlas

staged-coding-4-buildlisted

staged-coding stage 4 - build, the loop's return point. Test-first in vertical slices, fixes every call site of a shared helper, and stops after three failed attempts at the same defect. Signs its ledger line on finish. Triggers: /staged-coding-4-build, or invoked in order by /staged-coding.
Shadysmetools/claude-skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill Shadysmetools/claude-skills
# Stage 4 - build (the loop's return point) Part of the `/staged-coding` chain. Runs after stage 3 (plan) and stage 3.5 (pre-mortem) and before stage 5 (house style). **This is also the return point of the inner loop**: stage 8 (prove it), stage 8.5 (design match) and stage 9.5 (refuter) send you back here on a failure. Stages 8 and 8.5 share one cap of `config.max_prove_rounds` rounds before the run must stop and report honestly; a stage 9.5 kill always comes back here, and the refuter keeps attacking the repaired form. Runs on `config.builder_model`. The conductor carries the cross-cutting rules; read them there. ## Inputs - The plan from stage 3 (or its "skipped, mechanical" note) and the council verdict if one sat. - **Stage 3.5's failure-modes table**: each row with TEST? = N becomes one of the tests written first. - On a loop-back: what broke, what was tried, and the round count so far. ## What to do Test-driven development (use a TDD skill if you have one): one failing test -> minimal code to pass -> repeat. **Vertical slices only**: never write all the tests and then all the code. Each test responds to what the previous cycle taught you. On an EPIC, build every slice in this one run, slice by slice inside this stage. **Shared helpers:** if the fix lives in a shared/exported helper, grep **every** call site and fix or explicitly defer each one, not just the call site that surfaced the bug. ### Rules (a) **Three strikes.** Count fix attempts. After the THIRD fa