[stphung]

Summary

Ran the test-audit skill (3 parallel read-only agents across the full tests/*_tests/ scope) per the scheduled routine. One auto-fixable finding, several flagged for human review.

• Auto-fixed: tests/item_tests/item_pipeline_test.rs::test_roll_rarity_prestige_bonus_shifts_toward_higher_tiers had a Monte Carlo threshold set at only ~1.7 sigma below the expected mean (the test's own comment noted a ~5% false-failure rate). Increased trials 1500 → 8000 and widened the margin to ~3.2 sigma (false-failure rate now <0.1%).
• Flagged for human review (not auto-fixed, per skill rules):
  • HIGH: src/combat/enemy_generation.rs still calls rand::rng() directly (unlike items::drops, fixed in Fix unseeded RNG in item drop generation (test-audit) #692) — a production-code RNG-injection change, outside this skill's scope.
  • Several redundant/duplicate test pairs (soulforge discovery guard clause across 3 files, Deep pool-refresh 6h boundary across 2 files, 7 achievement-handler test pairs) — removing/consolidating tests needs human judgment.
  • The known ~40-60+ instance brute-force RNG-search loop backlog (tracked since 2026-07-02), plus a missing shared helpers.rs in tests/game_tick_tests/ — both previously flagged, still unresolved.

Full findings detail is in the history log entry added in this PR.

Verification

• cargo test --test item_tests test_roll_rarity_prestige_bonus_shifts_toward_higher_tiers — passes
• make check — fmt, clippy, full test suite (2908+3074+... tests), progression check, voyage simulator, and security audit all green
• cargo test run 10x consecutively — 0 failures across all runs (flakiness verification per the skill's Phase 3)

Test plan

• Targeted test passes with new margin
• make check passes
• 10x consecutive cargo test — 0 failures

🤖 Generated with Claude Code

https://claude.ai/code/session_01XAYN2B3RiSrxmVB6aNi8AL

---

Generated by Claude Code