lego-cubes-prep

name: lego-cubes-prep description: End-to-end preparation of a new Lego Cubes podcast episode (Pavel's Russian-language podcast about content + dev projects). Walks through 7 stages with check-ins — collect Content bullets from Tuclaw/Podcast files, fill the episode note, suggest an episode title in the established poetic style, find IMDB/Steam/Goodreads links, produce a LEGO-style cover image prompt for Nano Banana Pro, after the recording publish the .mp4 + .jpg + .nfo bundle to the NAS for Plex/Kodi, and finally archive the per-content impression files into References/. Trigger on any mention of preparing, assembling, filling, publishing, or archiving a Lego Cubes episode — including phrases like "prepare episode 30", "Lego Cubes 30", "the next episode", "fill out the episode note", "I recorded the podcast", "put it on the NAS", "make the nfo", "faststart the recording", "move impressions to References", "archive the episode", or a bare episode number with podcast context. Pavel typically writes these requests in Russian — same triggers apply regardless of language. Always use this skill when Pavel mentions a Lego Cubes episode by number or asks to prepare/finalize/publish/archive one — do not handle these requests ad-hoc, the skill encodes critical conventions (ASCII hyphens, sequel grouping, title style, single-room cover composition, NAS naming, nfo plot format, References archival rules) that are easy to miss without it.

Lego Cubes Episode Preparation

This skill prepares a new Lego Cubes podcast episode for Pavel. It walks through 5 stages with explicit check-ins so Pavel can review and adjust at each step before moving on.

Vault layout (always assume this structure)

• Vault root: /Users/pavel.karpovich/Obsidian/PK Workspace/
• Episode notes live at vault root: Lego Cubes NNN - <Title>.md (NNN is 3-digit padded)
• Per-content impressions (films/shows/games/books for the upcoming episode) live in: Tuclaw/Podcast/
• After an episode is published, those files get moved to References/ - that migration is NOT part of this skill, it happens later. We work only with what is currently in Tuclaw/Podcast/.
• Template: Templates/Lego Cubes Episode.md (Templater plugin) - creates a stub file Lego Cubes NNN -.md with frontmatter

Critical conventions (do not violate)

• ASCII hyphen - only, never — (em dash) or – (en dash). This is enforced project-wide. The user has rejected em dashes explicitly.
• Content bullets format: - <hook> - [[<wikilink with year>]]
• Sub-items for marathons/dilogies: indent 4 spaces, same hook format
• Frontmatter episode: 30 (integer, not padded)
• File on disk: Lego Cubes 030 - <Title>.md (padded to 3 digits)

The 7 stages

Stages 1-5 are pre-recording (Obsidian note prep). Stage 6 is post-recording publishing (.mp4 to NAS with sidecars). Stage 7 is post-publishing archival (move impression files to References/). Pavel often invokes only one stage at a time — phrases like "I recorded the episode, prepare the file" or "put it on the NAS" jump to Stage 6; "move impressions to References" jumps to Stage 7. Don't run earlier stages in those cases; just go to the requested one.

Run pre-recording stages in order. After each stage, summarize what was done and ask Pavel to confirm before moving on. Pavel may pause the skill at any stage - that is normal.

Stage 1 - Episode setup

Goal: locate or create the episode note.

1. Determine episode number N from Pavel's request. If unclear, ask.
2. Compute padded form: NNN = String(N).padStart(3, "0").
3. Check if a file matching Lego Cubes NNN - *.md exists at vault root. Use ls "/Users/pavel.karpovich/Obsidian/PK Workspace/" | grep -i "Lego Cubes NNN".
4. If file exists, Read it - confirm to Pavel which file you found and proceed.
5. If no file exists, tell Pavel to create one via Obsidian Templater (Cmd+P -> "Templater: Create new note from template" -> "Lego Cubes Episode") and give the episode number when prompted. Wait for him to do this and confirm. Re-check.
6. The template produces a stub with empty title:, date:, cover: and skeleton sections (## Logo, ## Links, ## Key notes -> ### Content, ### Projects).

Stage 2 - Collect Content

Goal: fill the ### Content section with one-line hooks for each piece of content discussed in this episode.

1. List files in Tuclaw/Podcast/: ls "/Users/pavel.karpovich/Obsidian/PK Workspace/Tuclaw/Podcast".
2. Filter to files belonging to THIS episode by their > Mentioned in [[Lego Cubes NNN -]] marker at the bottom. Use:

   cd "/Users/pavel.karpovich/Obsidian/PK Workspace/Tuclaw/Podcast" && for f in *.md; do mention=$(grep -o "Lego Cubes [0-9]\+ -" "$f" | head -1); echo "$mention | $f"; done | sort
   

   Pick rows matching the target episode number.
3. Read each matched file in parallel (single tool message with multiple Read calls - Pavel cares about latency).
4. For each file, extract a one-line hook:
   • Skip files without a review body - if the file has only frontmatter and the > Mentioned in line (no impression text), drop it. Files with status: watching/playing/on hold and empty body are typical skips. Tell Pavel which ones you skipped and why.
   • These hooks become the published show notes, so keep them short - a teaser, not a verdict. Aim for roughly 3-8 words, one phrase. Do NOT spell out the reasoning or reveal the whole point: a three-word reaction is a hook, a full sentence explaining why belongs in the review, not the notes. Pick one memorable angle Pavel emphasized, then stop.
   • Avoid loud, overblown claims - understate rather than oversell. Pavel has pushed back more than once when hooks ran too long or too loud; when in doubt, cut.
   • Keep hooks in Russian, in Pavel's voice (he writes informally).
5. Group related items under a parent bullet:
   • Sequels watched together (e.g., "Дилогия X перед выходом второго")
   • Book + film adaptations of same source
   • Genre marathons (e.g., "Резидент-марафон после RE9") Use 4-space indent for sub-items.
6. Order: movies -> shows -> games -> books (matches the established style in episodes 024-028). Sub-grouped items keep their parent in this order based on dominant medium.
7. Format every bullet as - <hook> - [[<wikilink>]]. The wikilink is the file name without .md (e.g., [[Project Hail Mary (2026)]]).
8. Show the proposed Content block to Pavel as plain text in chat first, before writing to the file. Ask if grouping/order/wording is OK.
9. After confirmation, Edit the file to replace the empty ### Content\n\n- placeholder (followed by ### Projects or end of file).

Reference - example Content block (from episode 029):

- Дилогия Devil Wears Prada перед выходом второго
    - Пересмотрел оригинал, добрые эмоции и отличная актёрская игра - [[The Devil Wears Prada (2006)]]
    - Идеальное продолжение про трансформацию медиа под новые реалии - [[The Devil Wears Prada 2 (2026)]]
- Странный фильм от режиссёра Drama про чужие сны и cancel culture - [[Dream Scenario (2023)]]
- Четвёртый сезон ощущается как промежуточный, но войну с Вилтрумитами показали круто - [[Invincible S04 (2026)]]
- DLC к ремейку RE4 за Аду, чисто сюжетная история без новых механик и локаций - [[Resident Evil 4 - Separate Ways (2023)]]

The 029 hooks above run a little long - they are kept only as a structure/grouping reference. Episode 030 deliberately trended shorter and is the better length target: look at the ### Content block in Lego Cubes 030 - *.md for the calibrated brevity. If a hook reads like a sentence lifted from the review, trim it to a phrase.

Stage 3 - Suggest title

Goal: propose 4-5 episode title candidates in the style of recent episodes. Pavel iterates with you until he picks one.

1. Read 5 most recent episodes to absorb the style. Use Glob to find them, then read 023-028 (or whatever's recent). The current style (022 onward) is poetic, metaphorical, 4-7 words, blending the central project of the period with a content theme. Examples:

   • 022 "Когда агент зовет в континуум" (Continuum project + agents)
   • 026 "Снежная кузница параллельных сеансов" (winter break + Continuum parallel execution)
   • 027 "Оруженосец из черепашьей гавани" (Tuclaw assistant + Turtle Ecosystem)
   • 029 "Дьявол носит артефакт повелителя теней" (Devil Wears Prada + Maul Shadow Lord + Glitch project as "артефакт" double meaning)

2. Inputs to the title:

   • The ### Content you just filled (look for dominant content themes)
   • The ### Projects section in the same file (Pavel fills this himself before invoking the skill - if it's empty, ask him to fill it first; the projects section drives the title's central image more than content does)

3. Brainstorm angles:

   • One central project name or metaphor as the anchor
   • One content reference as the secondary layer
   • Look for double-meaning words: "артефакт" (tech bug + magic relic), "фантом" (Star Wars + glitched data), "сбой" (malfunction + disruption), "трещина" (crack in mask + crack in code)
   • Avoid in-your-face references - Pavel prefers subtle and metaphorical over literal

4. Present 4-5 candidates with one-line reasoning each explaining what's woven in. Group them by angle (project-anchored vs content-anchored vs hybrid) so Pavel can quickly see the spread.

5. If Pavel likes the start but not the end (or vice versa), keep the part he likes and brainstorm only the variable half. Push toward double meanings and cinematic phrasing - Pavel responded well to those in past sessions.

6. After Pavel picks, instruct him to:

   • Rename the file: change Lego Cubes NNN -.md to Lego Cubes NNN - <Chosen Title>.md (in Obsidian, just rename the note - Obsidian will update wikilinks automatically)
   • Set title: <Chosen Title> in frontmatter (just the title, no episode prefix)

   You can offer to do this via Bash mv + Edit, but warn that doing it from the file system bypasses Obsidian's wikilink updater - safer to let Pavel rename in Obsidian. If no other notes link to this episode yet (likely the case for a freshly-created episode), mv is fine.

Stage 4 - Find links

Goal: fill the ## Links section with IMDB/Steam/Goodreads URLs for each Content item.

1. Walk through the Content items in the same order they appear in the section.
2. For each item, pick the right platform:
   • Movies, TV shows -> IMDB. URL format: https://www.imdb.com/title/ttXXXXXXX/
   • Games (including DLC) -> Steam. URL format: https://store.steampowered.com/app/XXXXXXX/<Slug>/. If a game is not on Steam (rare), use rawg.io as fallback.
   • Books -> Goodreads. URL format: https://www.goodreads.com/book/show/XXXXXXX-<slug>
3. Run WebSearch in parallel for all items - one search per item with the title and year, e.g. "Dream Scenario" 2023 IMDB. Spawn all WebSearches in a single tool message to keep latency low.
4. From the results, pick the canonical title page URL (not mediaviewer/cast/reviews subpages).
5. Edit the ## Links section, replacing the placeholder - with the bullet list. Maintain the same order as Content. No descriptions on the links - just bare URLs as bullets, matching the style of episodes 024-028.

Reference - example Links block:

## Links

- https://www.imdb.com/title/tt0458352/
- https://www.imdb.com/title/tt33612209/
- https://www.goodreads.com/book/show/51901147-the-ballad-of-songbirds-and-snakes
- https://store.steampowered.com/app/2109300/Resident_Evil_4__Separate_Ways/

Stage 5 - Cover prompt for Nano Banana Pro

Goal: produce a LEGO-style image prompt for Pavel to feed into Nano Banana Pro (Gemini 3 Pro Image). All Lego Cubes covers follow a fixed LEGO brick diorama aesthetic - the variables are camera, subject, palette, and Easter eggs.

Read references/nano-banana-cover-prompt.md before writing the prompt. That file has the full Nano Banana Pro prompting principles, the Lego Cubes house style spec (style anchor, color palette by LEGO brick names, minifigure rules, past episode style calibration), the composition decision tree (single-scene vs multi-zone), the prompt assembly template, and iteration guidance. Loading it lazily keeps this skill body lean.

Workflow:

1. Extract the central image from the chosen title - the cover should literalize the title's main figure or concept.
2. Decide composition (single coherent scene vs multi-zone diorama) using the decision tree in the reference file. Default single-scene unless Pavel says otherwise.
3. Build the prompt by filling the assembly template in the reference file. Include all required ingredients (subject, composition, action, location, style + lighting + text + aspect ratio).
4. Apply Nano Banana Pro best practices: prose over tag soup, explicit text rendering instructions, named LEGO brick colors, full sentences.
5. Weave in 2-3 subtle Easter eggs referencing other Content items - small props with named labels, posters, silhouettes. Not dominating the scene.
6. Output the prompt as a single fenced code block in chat (one-click copy), followed by a 3-5 bullet ## Что в нём заложено summary so Pavel can decide whether to tweak.
7. If Pavel asks to change direction, update only the affected block of the prompt - keep the style anchor and lighting boilerplate stable.

The cover image filename, once Pavel generates it, follows the pattern LEGO Cubes E029.jpg (or .jpeg/.png) inside the vault. Frontmatter cover: and ## Logo embed both reference this. Stage 5 stops at the prompt - generation and embedding are out of scope. Stage 6 picks the image back up to publish it on the NAS.

Stage 6 - Publish to NAS (post-recording)

Goal: process the raw OBS recording into a streaming-ready file with metadata sidecars (image + nfo) so Plex/Kodi index the episode. Run this after the recording session is done.

Naming convention on NAS (differs from the vault):

• Single-digit episode: LEGO Cubes E29.mp4, NOT zero-padded E029. Vault uses padded Lego Cubes 029 - <Title>.md for sortability, NAS does not.
• Per-episode files: LEGO Cubes E<N>.mp4, LEGO Cubes E<N>.jpg, LEGO Cubes E<N>-thumb.jpg, LEGO Cubes E<N>.nfo (all four required).

Steps:

1. Find the source recording. OBS writes to /Users/pavel.karpovich/Movies/YYYY-MM-DD_HH-MM-SS.mp4. Pick the largest file from today (small <100MB files are throwaway tests):

   ls -lt /Users/pavel.karpovich/Movies/$(date +%Y-%m-%d)*.mp4
   

   If multiple plausible candidates, ask Pavel which one is the real recording.

2. Remux with +faststart. This rewrites the file with the moov atom at the front so streaming/seek works without downloading the whole file. No re-encode (-c copy), so it runs at ~250-300x speed:

   ffmpeg -i "<source>.mp4" -c copy -movflags +faststart "/Users/pavel.karpovich/Movies/LEGO Cubes E<N>.mp4"
   

   Watch the ffmpeg output - capture these for the nfo:

   • Duration: HH:MM:SS.SS -> total seconds for <durationinseconds>, rounded minutes for <runtime>
   • Resolution (3840x2160 -> <width>3840</width><height>2160</height><resolution>4K</resolution>)
   • Codec (h264), audio codec (aac), audio channel count (stereo -> 2)

3. Copy files to NAS. The .mp4 is 15-25 GB, so run that one in the background while you do the rest:

   cp "<image-source>" "/System/Volumes/Data/mnt/nas/media/me/LEGO Cubes/LEGO Cubes E<N>.jpg"
   cp "<image-source>" "/System/Volumes/Data/mnt/nas/media/me/LEGO Cubes/LEGO Cubes E<N>-thumb.jpg"
   # mp4 in background - takes 5-10 minutes over the network mount:
   cp "/Users/pavel.karpovich/Movies/LEGO Cubes E<N>.mp4" "/System/Volumes/Data/mnt/nas/media/me/LEGO Cubes/LEGO Cubes E<N>.mp4"
   

   The two .jpg copies are byte-identical content; both names exist because Plex/Kodi look for either, mirroring the pattern of past episodes (E24+ have both E<N>.jpg and E<N>-thumb.jpg).

   The cover image source is usually ~/Downloads/Generated Image *.jpg from Nano Banana Pro - Pavel will paste a path or attach the image. If he attached an image, ask for the local path.

4. Build the nfo. Read the previous episode's nfo (LEGO Cubes E<N-1>.nfo) before writing - it is the canonical layout to mirror exactly. Source the title and plot from the vault note (Lego Cubes NNN - <Title>.md).

   Plot field formatting (this is what makes Stage 6 different from a simple copy-paste):

   • Sections, each labeled and separated by a blank line, in this order: General: (only if the vault note has a non-empty ### General), then Content:, then Projects:. Every bullet line gets a 2-space indent (matches E28's nfo style - E28 included all three sections; E29 had no General so its nfo had only Content + Projects). Do not drop General when the note has it - Pavel flags this.
   • Strip all [[...]] wikilink syntax. For Content bullets - <hook> - [[Title (Year)]], the link target becomes plain text: Title (Year). For Projects bullets that mention [[Tuclaw]], [[Glitch]], etc., drop just the brackets: Tuclaw, Glitch.
   • Sub-bullets (dilogies, marathons) keep their 4-space indentation under the parent. Add a trailing : on the parent line for readability (e.g., - Дилогия Devil Wears Prada перед выходом второго:).
   • Escape & as & in titles like The Hunger Games - The Ballad of Songbirds & Snakes (2023) (XML requires it).
   • Do NOT include the ## Links section, frontmatter, or anything else from the vault note. Just General (if present) + Content + Projects.

   Other nfo fields:

   • <title> and <originaltitle>: same value, the chosen episode title (no Lego Cubes prefix, no episode number).
   • <showtitle>LEGO Cubes</showtitle>, <season>1</season>, <episode>N</episode>, and <group episode="N" id="AIRED" name="" season="1"/> (note: <episode> integer, not padded).
   • <premiered> and <aired>: recording date as YYYY-MM-DD.
   • <dateadded>: same date with time YYYY-MM-DD HH:MM:SS.
   • Stream details: codec/aspect/width/height/resolution/durationinseconds from Step 2's ffmpeg output. Aspect for 3840x2160 is 1.78.
   • <original_filename>LEGO Cubes E<N>.mp4</original_filename>.

5. Verify all 4 files landed correctly:

   ls -la "/System/Volumes/Data/mnt/nas/media/me/LEGO Cubes/LEGO Cubes E<N>"*
   

   The .mp4 size on NAS must match the local source byte-for-byte (network copies can stall).

Stop after this stage. The local Movies copy stays as a backup; don't delete it. The original raw YYYY-MM-DD_HH-MM-SS.mp4 from OBS also stays - Pavel cleans those up himself later.

Stage 7 - Archive impressions to References

Goal: move the per-content impression files for this episode from Tuclaw/Podcast/ to References/. This happens AFTER publishing (Stage 6), when the impressions are no longer "upcoming material" but historical record.

What gets moved: only files with a written review body. Skip stubs (frontmatter only with status: watching/playing/on hold and no impression text under the > Mentioned in line). Those are works-in-progress that may roll over into a future episode - they stay in Tuclaw/Podcast/.

Steps:

1. List all files in Tuclaw/Podcast/ marked for this episode:

   cd "/Users/pavel.karpovich/Obsidian/PK Workspace/Tuclaw/Podcast" && for f in *.md; do
     grep -l "Lego Cubes NNN" "$f" 2>/dev/null
   done
   

   (Substitute the 3-digit padded episode number.)

2. Filter to filled-in files. A reliable heuristic: file size. Stubs are <500 bytes (~20 lines of just frontmatter + the mention line); filled impressions are 1500+ bytes. Verify with:

   for f in <matching files>; do echo "$(wc -l < "$f") lines | $(stat -f %z "$f") bytes | $f"; done | sort -n
   

   Anything under ~25 lines or 500 bytes is almost certainly a stub - Read it to confirm before deciding. Tell Pavel which ones you're skipping and why.

3. Cross-check against the published episode's ### Content section. Every Content bullet's wikilinked file should be in the move list; if a file is in Content but not in your move list, that is a bug - stop and investigate.

4. Move the filled files in one batch:

   cd "/Users/pavel.karpovich/Obsidian/PK Workspace" && for f in "<title 1>.md" "<title 2>.md" ...; do
     mv "Tuclaw/Podcast/$f" "References/$f" && echo "moved: $f"
   done
   

   References/ is flat (no subfolders). Filesystem mv is safe here because the basename is unchanged - Obsidian's [[Title (Year)]] wikilinks resolve by shortest unique name and survive the move without rewriting.

5. Verify - the moved titles should no longer appear in Tuclaw/Podcast/:

   ls "/Users/pavel.karpovich/Obsidian/PK Workspace/Tuclaw/Podcast/"
   

   Only stubs (and impressions for the NEXT episode) should remain. List the remaining stubs so Pavel can see what is queued for the next round.

6. Roll the stubs forward. Stubs marked for THIS episode obviously won't make it - they have no review body and the episode is already published. Update their > Mentioned in [[Lego Cubes NNN - <Title>]] line to point to the next episode in bare form: > Mentioned in [[Lego Cubes <NNN+1> -]] (trailing dash, no title - the next episode file doesn't exist yet, but the wikilink will resolve once Pavel creates it via Templater).

   Find stubs to roll: any remaining file in Tuclaw/Podcast/ whose Mentioned in line still names the just-published episode. Older episodes' stubs may also still be here (rolled forward repeatedly across episodes - that's fine, just bump them all to the next episode).

   cd "/Users/pavel.karpovich/Obsidian/PK Workspace/Tuclaw/Podcast" && for f in *.md; do
     grep "Mentioned in" "$f"; echo " | $f"
   done
   

   Then Edit each file's Mentioned line. Confirm afterward that every remaining stub points to the next episode.

End of skill. The episode is fully prepped, published, and archived; rolled-over stubs are queued for the next round.

Workflow philosophy

• Pace it like a checklist with check-ins, not a one-shot dump. Pavel always reviews the Content block before it lands in the file, always wants 3-5 title options to compare, and often iterates twice on the cover prompt. Bake those pause points in.

• Read multiple files in parallel. Stage 2 and Stage 4 both have list operations - issue all the Read or WebSearch calls in a single tool message. Pavel notices latency.

• When in doubt about stylistic conventions, look at episodes 024-028. Those are the canonical recent style. The earlier episodes (001-021) used different conventions and should not be mimicked for current work.

• Pavel writes in Russian, prefers informal voice. Don't formalize his hooks. Keep grammar but allow casual phrasing ("зашло", "крепкая восьмёрка", "не докрутили", "слили концовку"). Mirror his vocabulary from the source impressions.

• The skill stops after the cover prompt. Image generation, file uploads, publishing - all out of scope. Just hand off the prompt and confirm the episode note is fully populated.

Quick reference - file paths