← ClaudeAtlas

staged-coding-3.5-premortemlisted

staged-coding stage 3.5 - pre-mortem. Before any code exists, table every way the change can fail silently, plus rollback and migration safety. Signs its ledger line on finish. Triggers: /staged-coding-3.5-premortem, or invoked in order by /staged-coding.
Shadysmetools/claude-skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill Shadysmetools/claude-skills
# Stage 3.5 - pre-mortem (what breaks, and does anyone see it) Part of the `/staged-coding` chain. Runs after stage 3's plan and before stage 4 writes any code. Callable alone, but the conductor carries the cross-cutting rules (unit of work = the whole epic, lane, which model builds vs checks, the ledger law, stop conditions). Read those there. **Why this stage exists.** Stage 6 hunts swallowed errors AFTER the code is written. Nothing asks "what happens when this throws and nobody is watching" BEFORE. Lesson: one unguarded call on a null can blank a whole page while every check stays green. A pre-mortem is cheap before the code exists and expensive to reconstruct after it ships. ## Inputs - The plan from stage 3 (or, on a mechanical ticket that skipped 3, the ticket plus stage 1's traced call path). - The lane (BUG or EPIC) and the list of files/paths the change will touch. - Stage 1's live-state read (what the real data looks like). ## What to do ### 1. The failure-modes table - every touched code path gets a row | CODEPATH | FAILURE MODE | RESCUED? | TEST? | USER SEES? | LOGGED? | |---|---|---|---|---|---| One row per code path the change touches: the render expression, the API call, the DB write, the queue read. For each: what can fail (null, timeout, empty result, partial write, double-fire), is it caught, is there a test for that path, what the user sees when it happens, does it log. **Mechanical rule, no judgment: any row with RESCUED = N, TEST = N and USER SE