Trampelpfadanalyse — LLM-sichtbare Konventionen empirisch herstellen

ellmos-ai/trampelpfadanalyse

Error analysis for pipeline and control-file workflows: check whether a convention or procedure is actually visible and discoverable to an LLM. Empirical baseline → intervention → retest comparison using naive subagents (isolated sandbox copies, identical test case, quantitative success measurement). Use this skill when agents repeatedly ignore a rule/README/convention or navigate incorrectly, and you want to measure whether a documentation change actually changes the behavior. Triggers on "is t