psycho-baller/content-pipeline

content-pipeline

orchestrates the full 3-step content pipeline. accepts a single file, a list, a directory, or auto-detects unprocessed transcripts.

transcript source

/Users/rami/Library/Mobile Documents/iCloud~md~obsidian/Documents/Obsidian/My Outputs/Transcriptions/

output root

/Users/rami/Documents/life-os/notes-processing/

each transcript gets its own folder:

notes-processing/{filename}/
  research.md   ← step 1
  hooks.md      ← step 2
  content.md    ← step 3

invocation modes

single file:

"run content pipeline on my-note.md"

specific files:

"run content pipeline on file-a.md and file-b.md"

directory:

"run content pipeline on all files in /path/to/folder"

unprocessed (default):

"process unprocessed transcripts" / "run pipeline on new transcripts" scan the Transcriptions folder, find files where notes-processing/{filename}/content.md does not exist, process those.

preprocessing (apply before every step)

all transcripts contain Obsidian syntax that must be cleaned before processing:

• strip YAML frontmatter (everything between --- at the top)
• convert wiki-links: [[Note|display]] → display, [[Note]] → Note
• remove ## Related Notes section and everything after it
• work with clean plain text from this point forward

content length tiers

assess each transcript after cleaning and apply the appropriate depth:

| tier | word count | research depth | hooks per medium | |------|-----------|----------------|-----------------| | short | < 150 words | 3-5 sources, 1-2 claims | 3 per medium | | medium | 150-500 words | 5-8 sources, 3-5 claims | 4 per medium | | long | 500+ words | 10+ sources, all claims | 5 per medium |

steps per file

run in order:

1. content-research — clean transcript, extract claims by tier, do web research, write research.md
2. content-hooks — clean transcript, identify dominant themes, research trends, generate hooks + full tweets + ICE scores, write hooks.md
3. content-writer — clean transcript, pick top ICE hook, write master long-form piece infused with research, write content.md

print after each step: [{step}] done — {filename}

completion report

after all files are processed, output a summary table:

| file | tier | overall ICE | linkedin | twitter | ig/tiktok | youtube | |------|------|-------------|----------|---------|-----------|---------|

skipped files (already processed): list with "already done".

notes

• if a step's output already exists, skip it and continue from the next
• create output folders as needed
• if a specified file doesn't exist, say so and stop
• for scattered multi-topic transcripts, identify the 1-2 dominant threads and focus the pipeline there