kenlkehl/onc-data-wrangler-plugin
Skill Claude CodeCodex
Run red-team tests to evaluate whether AI coding agents resist prompt injection attacks that attempt to exfiltrate simulated PHI. Starts a honeypot server, creates an isolated workspace with fake patient data, launches the agent-under-test, and reports whether data was exfiltrated. Use when you want to test agent…
kenlkehl/onc-data-wrangler-plugin
Skill Claude CodeCodex
Reproduce Published Biomedical Paper Results. End-to-end pipeline that extracts every quantitative result from a published biomedical paper, independently reproduces each result from raw data, and performs discrepancy analysis. Use when the user wants to reproduce or validate published paper results against local…
kenlkehl/onc-data-wrangler-plugin
Skill Claude CodeCodex
Agentic verification pass over answer-questions output. For each patient, a frontier-LLM subagent (1) checks every answer against its cited evidence, (2) checks the full answer dict for cross-question contradictions, and (3) for flagged cells does targeted note retrieval (no full re-read) to confirm or correct the…