talk-roast

What problem does it solve? Before presenting a talk, speakers risk being caught on stage by wrong numbers, unsourced statistics, misattributed quotes, internal contradictions, and AI-sounding prose. This Skill performs an adversarial audit of a finished talk so those problems are found before the audience finds them. ## Core Features & Use Cases - Claim fact-checking: Extracts every number, quote, and named claim from the content brief, slides, delivery plan, and built deck, then verifies each against primary sources on the live web. - Seven-category hunt: Sweeps for untrue claims, internal contradictions, AI slop tells, easy errors like broken math, logic holes and Q&A landmines, reputation risks, and structural mismatches. - Severity-ranked report: Writes 06-roast-report.md with a verdict, a fact-check ledger, findings ordered from blocker to note, an AI-slop tally, and the three hardest hostile questions. - Use Case: After building a conference talk, ask for a brutal review and receive a report flagging that a slide cites a misattributed Einstein quote, a statistic with no traceable source, and a pie chart summing to 103%. ## Quick Start Roast my talk in the current working directory and tell me everything that will embarrass me on stage.