great_cto

Quality deepened, prompts on a diet, structured logs

• QUALITY-DEEPEN follow-ups #3–#6 (ARCH-quality-deepen-followups.md):
  archetype domain contracts extended 6 → 14 archetypes; contract regexes
  hardened against false positives (the /hold/-matches-"household" class);
  quality.mjs --gate --baseline (product-eval parity) and --trend
  (sparkline over metrics-history.jsonl); opt-in --browser harness with
  axe-core a11y + Web Vitals (LCP/CLS/INP) — dynamic playwright import,
  graceful na everywhere, runtime stays zero-dep.
• Agent prompt diet — architect.md 1181 → 827 lines, project-auditor.md
  1019 → 701 (−30% each): 16 embedded bash blocks extracted to scripts/*.sh,
  pricing/framework tables to skills/great_cto/references/, duplicate step
  numbering fixed, vestigial bd_create_if_new code removed. Cheaper every
  single agent invocation.
• Structured logging — zero-dep leveled logger (GREAT_CTO_LOG_LEVEL,
  ISO timestamps, error→stderr) across the board server; script diagnostics
  routed through it; hook protocol and CLI user output untouched by design.
• Board hardening — static path-traversal containment, origin-check dedupe,
  HOME-boundary for raw ?project= paths.
• Telemetry endpoint — default now the documented telemetry.greatcto.systems
  (verified live); personal placeholder hostname removed.
• Fixed the archetype-drift test failures (cli vs cli-tool) — main is fully
  green: lib suite 355, full suite 176, hooks 117.