What problem does it solve? Voice agent teams need to know which production calls failed, why, and how often, but manually reviewing hundreds of call transcripts is slow and error-prone. This Skill automates the triage of recent Cekura call logs against your KPIs, producing failure rates and a mutually-exclusive outcome breakdown with verbatim evidence. ## Core Features & Use Cases - Failure flagging with evidence: Identifies call IDs that hit each specified KPI or issue, quoting verbatim transcript slices and classifying failures against a taxonomy (loop, hallucination, workflow miss, ASR capture, and more). - Attribution-aware outcome distribution: Buckets every reviewed call into categories like not-answered, vetted, caller-side, or agent-issue, ensuring caller-side endings and recovered calls are not miscounted as agent failures. - Cost-efficient batch analysis: Uses cheap heuristics (call_ended_reason, duration, metric scores) first and only pulls full transcripts for ambiguous candidates, keeping token usage under control. - Use Case: Ask what percentage of the last 100 calls ended before all vetting questions were asked, and receive a report with per-issue failure rates, linked call IDs, and the share of answered calls that are improvable on the agent side. ## Quick Start Analyze the last 100 production calls for agent 1234 and tell me which calls failed to complete vetting and what percentage of answered calls are improvable.
