youre not overthinking it. adding too much fluff just increases the noise-to-signal ratio and makes it more likely to hallucinate [erroneous] logic. ive found that using a tree command output at the top of the file is way more effective than writing out descriptions of every folder. keep the instructions
extremely dense and focused on edge cases.
the more you explain, the more it ignores the actual logic . try moving the broad architectural context to a separate
ARCHITECTURE.md
and only keep the active implementation details in the claude file