Don't Start What You Can't Finish: A Counterfactual Audit of Support-State Triage in LLM Agents

Build an agent evaluation suite that implements the SSTA-32 diagnostic framework to test whether agents can identify blockers before executing. This addresses the common issue of agents blindly failing tasks they aren't equipped to complete.

Suggested repo: TriageEval

"Stop your agents from attempting the impossible—diagnose first, execute second."

Estimated effort: 30h