Skip to content

PROMPT.md

Latest commit

 

History

History
87 lines (64 loc) · 4.9 KB

PROMPT.md

File metadata and controls

87 lines (64 loc) · 4.9 KB
Raw

Programmarr — LLM Prompt

Copy everything below the line and paste it into your LLM (Gemini, Claude, ChatGPT, etc.).

Model recommendation: Use the largest, most capable model available — Claude Opus, Gemini Pro/Ultra, GPT-4o, etc. Speed-optimized models (Flash, Mini, Lite variants) tend to hit output limits or produce incomplete results on a task this large. This is one of those cases where the bigger model is worth it.

Set {TARGET} before you paste. Replace {TARGET} in the Rules section below with your desired channel count (e.g. 40). Too low and the LLM underfills the number blocks; too high and it pads with thin channels. A good rule of thumb: ~1 channel per 15–20 titles in your library.

Option A — File attachment (recommended if your LLM supports it): Paste the prompt, then attach plex_library.csv as a file. The LLM reads the CSV as structured data rather than raw text, which produces more accurate results and uses the context window more efficiently.

Option B — Single paste (works everywhere): Paste the prompt, then paste the full contents of plex_library.csv directly after it. Works with any LLM or interface that doesn't support file uploads.

Note on the Channel Numbering Scheme below: the category order shown is the default. When you copy the prompt from the web app, it regenerates that section to match your configured category order (Settings → Channel Numbering) and start number. The static text here is what the CLI uses.

You are a TV channel programmer. I have a self-hosted media server with the library listed below in CSV format. Your job is to design a set of themed virtual TV channels using only content from this library.

Rules

  1. Only use titles that appear in the provided CSV. Do not invent or suggest titles that are not in the list.
  2. A title can appear on multiple channels — a 90s comedy film should appear on the Comedy channel, the 90s Movies channel, AND the 90s Comedy Movies channel if they exist.
  3. Not every title needs a channel. Only include a title on a channel if it genuinely fits.
  4. Target approximately {TARGET} channels total. Aim for quality over quantity — a channel needs enough content to feel alive (at least 3–5 items for movies, at least 50 episodes for a TV marathon).
  5. Suggest new channels if you see clusters of content that would make a great themed station (e.g., if the library has 8 Pixar films, suggest a Pixar channel).
  6. Flag orphaned content — at the end, list any notable titles that didn't fit any channel. If 5 or more orphaned titles share a theme, suggest a new channel for them.

Channel Numbering Scheme

Number channels sequentially starting at 1, tight-packed in this category order — no gaps, no reserved ranges. A category with no channels is simply skipped (it consumes no numbers).

  1. TV Marathons — 24/7 single-show loops (needs 50+ episodes to qualify)
  2. TV Blocks — themed multi-show rotations (era blocks, genre blocks, etc.)
  3. TV & Movie Mix — mixed-genre channels spanning shows and films
  4. Movie Channels — genre and decade-based pools
  5. Studios / Directors / Actors — channels curated by creator or studio
  6. Networks — all shows from a single TV network
  7. Classic TV Blocks — historical lineups (TGIF, Must See TV, etc.)
  8. Franchise & Series — ordered collections (film series in release order, etc.)
  9. Specialty — single-movie loops, holiday, niche themes

Assign every channel the next free number — e.g. 12 marathons take 1–12, then TV Blocks start at 13.

Shuffle Types

Each channel must have one of these shuffle types:

  • ordered — plays content in strict order (use for franchises, chronological series)
  • shuffle — random rotation (use for genre pools, decade channels)
  • block — round-robin between shows/movies (2 episodes per show per turn; use for multi-show TV blocks)

Output Format

Output one channel per line as a JSON object (JSONL). No wrapper object, no markdown fences, no commentary between lines — just one {...} per line.

{"number": 1, "name": "Breaking Bad Marathon", "shuffle": "ordered", "content": ["Breaking Bad"]}
{"number": 2, "name": "Crime TV Block", "shuffle": "block", "content": ["The Wire", "Ozark", "Justified"]}
{"number": 3, "name": "80s Action Movies", "shuffle": "shuffle", "content": ["Die Hard", "Lethal Weapon", "Predator"]}

Each line must have exactly these fields:

  • number — integer channel number following the scheme above
  • name — string channel name
  • shuffle — one of ordered, shuffle, or block
  • content — array of title strings copied exactly from the CSV

After the channel list, you may add a brief plain-text note listing any notable orphaned titles and why they didn't fit.

The Library

(Option A: attach plex_library.csv as a file and delete this section) (Option B: paste the full contents of plex_library.csv here)