A disappointing output has exactly one of four root causes, and each one traces back to a concept you already have:
Under-specification. The prompt itself was missing a needed component — task, context, format, constraints, or examples (Module 2). The fix lives in the prompt.
Context overload. The conversation's context budget is strained (Module 1) — too much accumulated history, retrieved content, and instructions competing for the same space, producing degraded recall, repetition, or drift. The fix isn't a better prompt; it's a fresh session, a summary, or moving durable material into a Project.
Wrong feature or model. The setup itself doesn't fit the task — wrong entry point, missing feature, wrong model tier (Module 1's selection decisions). No amount of prompt rewriting fixes a tool mismatch.
Stale configuration. The Project's persistent instructions or knowledge sources have quietly drifted out of date (Module 5) — the exact "quiet risk" framing from that module, carried over here without modification.
The discipline is elimination, not guessing. These four can produce similar-looking symptoms. "Claude seems to be ignoring what I told it" could be under-specification (you never actually said it clearly), stale configuration (you told an earlier version of the Project, which has since drifted), or context overload (it got pushed out by everything since). The skill is checking which one actually applies, not defaulting to whichever explanation comes to mind first — because the fix for one does nothing for the others.