Scope Cursor's chat context with @-symbols instead of hoping the index finds the right files

Why this works

Cursor's default behavior when no explicit @-reference is given is to run a semantic search over the codebase's embedding index and surface whatever ranks closest to the question's phrasing, which works well when the relevant code happens to use similar vocabulary to the question and works poorly when it doesn't — a tax bug described in terms a reviewer would use ("charged twice") may rank a logging or analytics file that happens to mention "charge" more literally above the actual tax-calculation file that never uses that word. Naming @Files and @Folders explicitly removes that dependency on vocabulary overlap entirely and replaces it with a guarantee: the model reasons from exactly what's referenced, not from whatever an embedding-similarity ranking decided was close enough. @Docs and @Git matter for a different reason — they're sources the default semantic search over the codebase's own files wouldn't surface at all, since indexed documentation and commit history live outside the file tree the standard search operates over, so a question that genuinely depends on "what does the third-party API's field actually mean" or "when did this behavior change and why" needs those referenced on purpose or the model will answer from its own general knowledge of the API instead of this specific integration's documented behavior. SOURCE_JUSTIFICATION stated explicitly matters because it makes exclusion a deliberate act the model can be held to — without it, there's no way to tell whether the refund flow was left out because it's genuinely irrelevant or because it just didn't come to mind, and the rule against treating an unreferenced-but-similar file as equally authoritative closes the gap where the model reads one hop beyond what was referenced, finds something plausible-looking, and quietly reasons from it as if it had been included on purpose.