Updated project from new folder (UI + scenarios + features)

fix(grader): fake patch protection — reject patches < 20 chars with score 0.02, prevents random guessing

feat(scenarios): seeded variant selection — same seed always picks same variant, fully reproducible

feat(engine): InfraMindEnv — seeded reset, judge mode, episode trace export, feedback learning, FeedbackLearner, performance metrics

feat(base): seeded RNG, before/after metric scoring, root cause attribution, adversarial agent, failure report, skill breakdown, episode trace

feat(models): add MetricSnapshot, SkillBreakdown, FailureReport, EpisodeTrace, adversarial_hint, metric_improvement reward

feat(engine): AegisSwarmEnv — agent memory, dynamic difficulty, leaderboard, run history, feedback loop, RLock thread safety

feat(scenarios): custom scenario support + scenario registry — user-defined bugs become playable episodes

feat(task5): Auth Bypass security scenario — JWT none algorithm vulnerability, normal metrics, security-only signals

feat(task4): CPU Spike scenario — infinite recursion in sanitize(), no depth limit, 100% CPU. Hidden tests for MAX_DEPTH + WeakSet

feat(task3): Cascade Failure scenario — 3 variants: Redis timeout, retry storm, connection pool exhaustion. Signal vs noise

feat(task2): DB Deadlock scenario — 3 variants: lock ordering, TOCTOU race, N+1 transaction loop. Butterfly effect on restart

feat(task1): Memory Leak scenario — 3 variants: unbounded cache, event listener leak, unclosed DB connections

feat(variants): 9 randomized bug variants across 3 tasks — different bug every reset, no two runs identical

feat(scenarios): base scenario engine — butterfly effect, chaos injection, PR review, CI/CD, Jira, agent comms, safety layer

feat(models): typed Pydantic models — Action, Observation, Reward, SystemMetrics, Alert, AgentMessage