tend-perf

Tend Perf — the Quickener

Find the single highest-priority performance issue in this repo, fix it minimally, verify, and hand the diff back for review. Every fix must be mechanically provable from code inspection alone — no profiling, no benchmarks, no intuition. One issue per run.

0. Load config

• Read .claude/tend/config.yaml. Use skills.perf (off-limits globs, notes) + the modules: list.
• No config? Run at core tier: core SvelteKit + TS assumptions, no modules, conservative built-in off-limits (never touch lockfiles, CI, .env, config). Emit once: "run tend-onboard to sharpen." Continue.
• Active modules gate which reference packs load in Phase 2 (progressive disclosure). This skill only ever loads data and realtime — the other three (auth, validation, feature-flags) have no perf-specific pack and are ignored even if active.

1. Triage

MANDATORY READ references/scan-core.md.

Establish: active modules (from config), off-limits set (config globs + built-in rails), and this repo's hot paths — high-frequency event handlers (scroll/input/resize/WS/SSE), request handlers, and any loop over a collection that scales with real data (sessions, players, questions).

2. Scan — priority order, stop at the first real hit

Core (always): O(n²) algorithms with a clear O(n) alternative; redundant calculations or allocations inside hot loops or per-request middleware; missing debounce/throttle on high-frequency DOM events; repeated $derived filters re-traversing the same reactive array; images missing loading="lazy" or served uncompressed.

Module checks — load a module's reference pack only if its module is active:

• data → references/module-data.md — redundant DB calls executing on every request, getAll()/large-array deserialization for metadata-only lookups
• realtime → references/module-realtime.md — missing debounce/throttle specifically on WS/SSE tick handlers