thalarch-doubtlisted

# Thalarch Doubt Gate Final review is sometimes too late. A wrong architectural or causal decision becomes expensive after several files depend on it. This skill introduces a **bounded in-flight disconfirmation gate** for important decisions while correction is still cheap. Use doubt to reduce premature closure, not to manufacture hesitation. ## 1. Trigger only on meaningful decisions Activate when at least one applies: - branching/state behavior changes; - a module/service/data boundary changes; - correctness depends on thread safety, idempotency, ordering, lifecycle, ownership, or another property the compiler cannot prove; - a version-sensitive external API or framework pattern is load-bearing; - the codebase is unfamiliar and a wrong assumption would propagate; - the decision is difficult to reverse or has security/data/production blast radius; - a D2+ reasoning step has a strong leading answer but weak disconfirming evidence. Skip for mechanical renames, formatting, obvious one-line edits, pure inspection, or deterministic tooling operations. ## 2. Materialize the decision Write a compact decision artifact: - `DECISION` — what is about to stand; - `CONTRACT` — observable requirements/invariants it must satisfy; - `EVIDENCE` — only current facts supporting it; - `UNCERTAINTY` — what could still make it wrong. If the decision cannot be stated compactly, return to planning/reasoning before challenging it. ## 3. Preserve reviewer independence When a clean speci