{"slug":"ellmos-ai-trampelpfadanalyse","source_name":"ellmos-ai/trampelpfadanalyse","name":"Ellmos AI/Trampelpfadanalyse","description":"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","version":6,"lift":{"pass_rate_delta_pts":36.36,"pass_rate_pct":81.8,"total_cases":22,"passed_cases":18,"tokens_delta_pct":106.2,"turns_delta_pct":0,"verdict":"mixed","benchmark_model":"gemini-3.6-flash","grading_method":"judged","completed_at":"2026-08-06T17:41:33.838262+00:00"},"skill_score":0.8182,"benchmark_models":[{"model":"gemini-3.6-flash","headline":true,"delta_pts":36.36,"with_pass_pct":81.8,"without_pass_pct":45.5,"tokens_delta_pct":106.2,"turns_delta_pct":0,"total_cases":22,"cases_aggregated":20,"verdict":"mixed","never_hurt":false,"completed_at":"2026-08-06T17:41:33.838262+00:00","run_id":"5ab114d0-5526-4e9a-8d69-a38293e59ffa","version_number":6,"is_latest_version":true,"gate":null}],"trust":{"skill_safety":"passed","safety_status":"clean","intent_verdict":"safe","content_status":"clean","indexable":true},"license":"MIT","install_count":0,"manifest_hash":"4e1e1724d472427e58cf5a56950b78ff1394a6ef54b93306497805eb202475e0","raw_url":"https://app.decimal.ai/s/ellmos-ai-trampelpfadanalyse/SKILL.md","scorecard_url":"https://app.decimal.ai/skills/ellmos-ai-trampelpfadanalyse"}