← ClaudeAtlas

staged-coding-2.5-asklisted

staged-coding stage 2.5 - ask the owner what is genuinely unclear (blocking). One batched question set, before any code is written, about product decisions and ticket ambiguities only - never technical unknowns the codebase can answer. Signs its ledger line on finish. Triggers: /staged-coding-2.5-ask, or invoked in order by /staged-coding.
Shadysmetools/claude-skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill Shadysmetools/claude-skills
# Stage 2.5 - ask the owner what is genuinely unclear (BLOCKING) Part of the `/staged-coding` chain. Runs after Stage 2 (restraint) and before Stage 3 (plan). Callable alone (e.g. to interview the owner about someone else's PR before it is built further); the conductor carries the cross-cutting rules. ## Inputs - The ticket / slices read together in Stage 1 - The build / will-not-build list from Stage 2 - `config.project_owner`: the person whose decisions this stage collects ## Why this exists Without it, the pipeline's escape hatches all fire mid-run, on the agent's own judgment, after building has started: self-grilling interrogates the agent, not the owner, and a `NEEDS_CONTEXT` self-report at the end is easy to soften. Nothing turns to the owner before work begins. The failure it prevents: a slice built exactly as written can contradict itself (e.g. a picker whose headline list is people who cannot open the thing being sent to them), and a run can quietly make a real product call (e.g. "new records only") that the owner never made. ## What to do 1. **Read every slice together and write down each genuinely open item.** Classify before asking: - **Product decision**: only the owner can answer (who sees what, what happens to existing data, whether a behaviour is acceptable). **ASK.** - **Ambiguity in the ticket**: the words admit two readings that lead to different code. **ASK**, naming both readings. - **Technical unknown**: the codebase can answe