Find system failure modes and guardrails

Find unhandled failure modes and architectural blind spots

Clinical/healthcare software teams where data corruption and silent wrong results carry regulatory or safety risk.

• · Clinical Data Derivation Engine (CDDE) codebase

• · 3-pass review (runtime, cybersecurity, production readiness)
• · Prioritized findings: Critical/High/Medium/Low/Positive
• · Per finding: severity, trigger, current behavior, impact, suggested fix
• · File paths and line numbers for each issue

• · Subagents (WhatCanGoWrong, Cybersecurity, ProductionReadiness)
• · Code review (pattern matching on Python, asyncio, PydanticAI)

• · CDDE codebase structure understood (src/agents/tools.py, orchestration pipeline)
• · Subagents can run in parallel

• · Subagent timeout — falls back to single-pass review
• · Code patterns unfamiliar — produces generic findings (not specific fixes)
• · No test coverage documented — can't validate failure mode recovery

• · Structured pass: runtime → security → scale (defense-in-depth order)
• · Concrete examples: BEAST/POODLE/Sweet32 vuln checks, asyncio.gather error swallowing
• · Separates severity (critical/high) from nice-to-haves (low)