What problem does it solve? Reviewing an HTML slide deck by eye tends to stop at "looks good", while self-graded review loops either approve too early or oscillate forever. This Skill defines a strict, convergent review process that only declares a deck done when deterministic checks pass and fresh reviewers stop finding verified issues. ## Core Features & Use Cases - Deterministic hard gates: Measures text overflow, overlap, broken images, console errors, banned characters, contrast, and font loading via browser probes, with no LLM judgment involved. - Adversarial review loop: Dispatches fresh judge subagents per dimension (narrative, clarity, visual, delivery), then adversarially verifies every finding against a written constitution before fixing. - Convergence with a ledger: Tracks rejected findings so they are never re-litigated, terminates after two dry rounds or a round cap, and writes a per-round report with honest verdicts. - Use Case: You have a finished HTML deck built with a deck-stage engine and want it polished before a talk; the Skill serves it over HTTP, screenshots every slide, runs the gates and review rounds, fixes confirmed issues, and reports the evidence for the final verdict. ## Quick Start Review and polish my existing HTML slide deck until it passes all hard gates and two consecutive review rounds find nothing new.
