ずんだもん × 四国めたん narrate your coding-agent plans as auto-playing presentations (Claude Code plugin, VOICEVOX)
Agent
Blind first-time-viewer review of a zunda-presenter dialogue script. Spawn with ONLY the npm run view-deck -- --dialogue output as the prompt — no deck title, no topic, no context, no hint of expected findings (the review questions live here, not in the prompt). Returns unintroduced terms, thread losses, confusable…
Agent
Structure review of a deck outline BEFORE any dialogue exists. Spawn with the outline.md path, the structure.md reference path, and a one-paragraph framing of what the deck explains (audience and goal) — never the source material itself. Returns per-beat structural findings plus coverage questions derived from the…
Agent
Audits VOICEVOX engine readings for an assigned range of a zunda-presenter deck's lines. Spawn several in parallel, each with the FULL npm run readings output, the deck's dictionary.json content, the plugin root and deck dir paths, and an assigned line range. Verifies suspicions with try-reading and returns confirmed…
Agent
Visual layout check of a zunda-presenter deck's rendered slides. Spawn with the plugin root and deck dir paths (plus slide ids to restrict, if any); it runs npm run snap itself, reads every screenshot against the script's intent, and returns per-slide layout defects. Read-only apart from the snap output; never edits…