CE (Content Editing)

Last updated February 18, 2026

What This Does

Processes timestamped conversation transcripts and generates structured output for publishing.

Input

A timestamped transcript of a conversation (paste or file path).

Podcast Profiles

Different podcasts have different output formats and idea-types worth capturing. Always specify the podcast when invoking the skill (e.g. /ce tgs, /ce nonzero, /ce kosmo). If no podcast is specified, ask which one before proceeding.

Kosmopolitika

Ideas to look for (non-exhaustive):

  • Thesis statements — explicit claims or arguments being made
  • Framings/metaphors — new ways of seeing something (the ship dissolving, pregnant teenagers)
  • Distinctions — lines drawn between concepts (natural vs synthetic, good vs evil attractor)
  • Questions raised — unresolved tensions or open inquiries
  • Connections — unexpected links between domains (Gurdjieff ↔ Vernadsky)
  • Neologisms/coinages — new terms introduced (alienism, bio-aliens, sky aliens)
  • Reversals — moments where conventional wisdom gets inverted
  • Emotional peaks — where energy, vulnerability, or stakes feel highest

Signal phrases: When interlocutors say things like "this is interesting," "I like this," "that's funny," or similar — this is a strong indication that an idea worth capturing is being discussed. Pay attention to these moments.

Ideas from previous episodes will be collected in a separate archive for cross-referencing.

Output:

  1. Title and Thumbnail Idea — a compelling title + visual concept for the thumbnail
  2. One-Paragraph Description — concise summary suitable for show notes or social media
  3. References (with timestamps) — all people, books, movies, events, concepts mentioned, with precise timestamps for every occurrence [HH:MM:SS] Reference — brief context [HH:MM:SS] Reference — (second mention, if different context)
  4. Key Ideas (with timestamps) — core arguments, insights, themes with precise timestamps ``` [HH:MM:SS] Idea name or short label Brief description of the idea.
  5. [HH:MM:SS] Development or elaboration (if idea evolves) ```
  6. Clip Suggestions — segments worth extracting, organized by length (short under 2 min, medium 10-20 min, long 30-45 min)
  7. Cold Opens — see Cold Opens section below

Nonzero

Output differs from default. When /ce nonzero is invoked, produce this instead:

  1. Five title + thumbnail ideas. Each thumbnail must include 4-5 words of text that complement (not repeat) the title.
  2. Tweet-length description. One to two sentences, suitable as a tweet.
  3. Topic timestamps (table of contents). 6-12 timestamped topic shifts — not necessarily "ideas," but a navigational guide through the conversation. Each entry is a single phrase — descriptive of the content and intriguing enough to click on. No description or second sentence needed.
  4. Overtime marker. Clearly mark with "Going into Overtime:" where overtime is first mentioned. Follow it with what's discussed between that mention and the actual start of overtime. Then list overtime topics under "Discussed in Overtime:" as concise bullet points (no timestamps) — keep each to a short phrase.
  5. Cold Opens — see Cold Opens section below.
  6. No references list, no key ideas list, no clip suggestions.

TGS (The Glenn Show)

This is editing, not annotation. TGS CE produces an edit plan — editorial surgery on a 2-hour livestream to extract multiple video products. No references list, no key ideas list, no clip suggestions.

Workflow: 1. Nikita may feed live notes during the stream (editor notes, camera choices, segment observations). Remember these — they inform the edit plan. 2. Read the .txt transcript for overall comprehension and structure. 3. Use the .vtt for precise timestamps (grep searches — the file is too large to read in one pass). 4. Work in iterative passes, segment by segment. 5. Editing analysis first, titles & thumbs after.

Output — an edit plan (.md) with:

  1. Global editor notes. Camera instructions, color correction, name tags — anything that applies across all videos.

  2. Multiple video products. A 2-hour stream typically yields 3-4+ videos:

  3. Main segments (e.g., 45-min lecture discussion)
  4. Guest appearances (full segment + thematic extractions)
  5. Short clips for social (6-8 min)

  6. Note overlapping material between products.

  7. Per video: cold open, edit map, estimated runtime, then titles & thumbs (after all videos are mapped).

  8. Cold opens. Timestamp + verbatim quote. Note whether each option hooks only or hooks AND orients. See Cold Opens section below for principles.

  9. Edit map. A flat list of entries, each with in/out timestamps:

  10. KEEP — include this block. Say what's in it and why it matters.
  11. CUT — remove this block. Say what's in it and why it goes.
  12. TRIM — keep the core, cut the fat. Specify what to keep.
  13. YOUR CALL — could go either way. Make the case for and against.
  14. Every gap must be explicit. No implicit jumps between KEEPs. If there's dead air, a tangent, or transition between two KEEPs, mark it CUT.

  15. No table row/column cross-references — flat list only.

  16. Titles & thumbs. 5 pairings per video. Thumbnail text (4-5 words) complements the title, doesn't repeat it. Generated after the full edit map is complete.

  17. Summary table. At the end: video name, estimated runtime, one-line description of material.

Save to: ~/Documents/Projects/TGS/TGSL/YYYY-MM-DD-TGSL-XX/YYYY-MM-DD-edit-plan.md (XX = guest initials, e.g. CV for Chloe Valdary)

Methodology

Timestamp accuracy is critical. Follow this two-step process for all timestamp-dependent output:

Step 1: Content Pass

Read the full transcript and identify: - All references (people, books, movies, events, concepts) - Key ideas (arguments, insights, themes, metaphors, framings) - Potential clip boundaries and cold open candidates

Do NOT assign timestamps during this pass.

Step 2: Timestamp Pass

For each reference or key idea identified: 1. Search the transcript for the exact string (name, term, distinctive phrase) 2. Extract the timestamp directly from the VTT cue preceding the match 3. List ALL occurrences — if a concept appears multiple times, report all timestamps

Never estimate or approximate timestamps. Every timestamp must come from an exact string match in the transcript.

For key ideas: Search for the distinctive phrase or sentence where the idea is articulated. If an idea develops across multiple moments, timestamp each development point.

Cold Opens (all projects)

Propose short segments (ideally under 15 seconds each, max ~30 seconds) that could work as cold opens. Provide a few options for the main episode and a few for overtime (as teasers). Include timestamps and the verbatim quote.

What makes a good cold open: A cold open needs to do two jobs: hook (stop someone mid-scroll) and orient (give a sense of what the episode is about). A punchy 10-second zinger hooks but doesn't orient. A 45-second monologue orients but doesn't hook. The sweet spot is a passage that does both — vivid language, clear stakes, and enough framing that a new viewer understands the territory.

Principles: - Feature both speakers if possible — it gives the viewer the dynamic immediately - Prefer passages with vivid imagery or unexpected phrasing over abstract/cerebral ones - The best cold open candidate is often a passage where one speaker builds a clean arc in ~20-25 seconds: setup → image → stakes - Side threads and rambly tangents weaken momentum even if the content is interesting — cut anything that meanders - When proposing options, note which ones hook-only vs. hook-and-orient — let the editor choose the tradeoff

Quick Commands

  • process → Run full analysis on provided transcript
  • refs → Just extract references with timestamps
  • ideas → Just extract key ideas with timestamps
  • clips → Just suggest clips
  • title → Just generate title/thumbnail ideas