otaku-reco_skill

Overview

This skill is an anime recommendation and appreciation agent that generates compact, high-quality watch lists using real-time AniList data. It extracts user taste, runs targeted candidate queries with a local CLI script, and returns 6–10 curated picks with reasons and explicit caveats.

How this skill works

The agent first infers user preferences (mood, hard filters, era, episode limits, art style, and seed titles). It then calls a local Python CLI to fetch candidates either via a similarity query (seed-based) or a free-text search. Finally, it re-ranks results artistically and outputs grouped recommendations with AniList links, year, episode count, watch-order advice, and explicit 'watchout' notes.

When to use it

• You want a short, precise anime watchlist tailored to your mood or filters.
• You have a seed title and want close, explainable similar picks.
• You need quick choices with explicit caveats like slow-burn or ecchi content.
• You prefer recommendations based on live AniList metadata without an external database.

Best practices

• Provide at least one seed title or clear mood words for tighter results.
• Specify hard filters up front (no isekai, no harem, episode cap) to avoid unsuitable picks.
• Accept a first-pass suggestion; the agent can refine without asking more questions.
• Use the provided watch-order advice for multi-format franchises.
• Trust the explicit 'watchout' notes for pacing/content flags.

Example use cases

• A user who enjoyed 'Your Lie in April' asks for 6 shows with a healing/drama vibe and <=24 episodes.
• A user says 'similar to Steins;Gate' and receives a seed-based similar-list with why each title matches.
• A viewer wants short action series (<=12 eps) with little fanservice and gets concise alternatives.
• Someone unsure of taste uses mood words like 'suspense + sci-fi' and gets grouped recommendations.