Find system failure modes and guardrails
review-chaoscommandsetup L2★1
matt-grain/pharma-derive ↗What it does
Find unhandled failure modes and architectural blind spots
Best for
Clinical/healthcare software teams where data corruption and silent wrong results carry regulatory or safety risk.
Inputs
- · Clinical Data Derivation Engine (CDDE) codebase
Outputs
- · 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
Requires
- · Subagents (WhatCanGoWrong, Cybersecurity, ProductionReadiness)
- · Code review (pattern matching on Python, asyncio, PydanticAI)
Preconditions
- · CDDE codebase structure understood (src/agents/tools.py, orchestration pipeline)
- · Subagents can run in parallel
Failure modes
- · 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
Trust signals
- · 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)