[TimoKuronen]

Summary

• Distinguish tool/wiring failures from model review verdicts via AuditOutcome (exit 0 passed, 1 review required, 2 failed); skip critique persistence on tool failure
• Move sniff-system rule selection into AuditEngine.LoadSystemRulesAsync and centralize path validation; batch mode shares CLI error handling with sniff
• Inject HttpClient into OllamaAiOrchestrator with fake-handler unit tests; sort rule and target enumeration for deterministic prompts

Test plan

• dotnet test on CodeSmellAuditor.slnx (Release): 79 passing
• Exit 0 when every audited target reports Status: COMPLIANT
• Exit 1 when any target reports Status: REVIEW REQUIRED and none are tool failures
• Exit 2 when Ollama is unreachable or returns an empty/error stream, and no new critique file appears under WorkstationStorage/Reports/
• Exit 2 with a clean ERROR line (no unhandled crash) for a missing --stack folder in batch mode
• sniff-system still loads Rules/Architecture/ plus optional --stack packs from Rules/Stacks/