How Gigi Kenneth turns one Riverside podcast episode into a content pack with parallel agents

The problem

In her words: “Recording an episode is the easy part. What eats the afternoon is everything after”. That means writing a title that fits in 100 characters, a description with timestamps, tags, a Spotify blurb, a blog post and four or five social posts. She says “the grind is where voice drifts into generic AI slop.” She wanted it done in one pass, in her voice, “without me babysitting eleven separate prompts.”

The blueprint (as she describes it)

1. Set up once. Clone the repo and drop the skill folder into your Claude Code skills directory. Fill in a single show-profile.yaml with the show name, host, links, voice rules and the repo where packs get filed. If you leave a field blank, that deliverable is skipped. There is no code to touch.
2. Edit (optional, browser editors only). The skill tightens the episode inside Riverside’s web transcript editor. It drives Chrome and cuts tangents and production artifacts by clicking words in the on-screen transcript. It does not work with desktop apps like Premiere or Descript desktop.
3. Captions. It cleans the exported .srt. It strips hesitation markers and fixes mis-heard names, brands and jargon. Every timestamp stays intact.
4. Content pack (the core). It reads the show profile and the clean transcript once. Then it spawns one background agent per deliverable in the same turn, and all of them run concurrently. Every agent gets the same voice rules and guest-framing rules. The outputs include:
   • YouTube title options with character counts
   • a templated YouTube description
   • tags ordered highest-value first
   • a Spotify-safe description
   • a Markdown blog post with an FAQ for answer engines
   • LinkedIn, Instagram and TikTok posts
   • a YouTube community post
   • a newsletter blurb
   • pull-quotes for quote cards
   • chapter timestamps
5. File. It commits the whole pack to your content repo, one folder per episode, so past episodes are easy to diff.

You can run all four phases or just the one you need. Phases 2 to 4 are editor-agnostic.

The caption cleaner also works on its own as a dependency-free Python script:

python3 skill/scripts/fix_srt.py episode.srt --map corrections.json
# writes episode-corrected.srt

The map is a JSON file of wrong phrase to correct phrase. Short tokens that must not match inside other words go under a __word_boundary__ key. If you run it with no map, it only strips the hesitation markers.

Stated results

None stated. The post says eleven pieces get drafted at once, but it gives no time-saved, reach or usage figures.

Why this matters for postby.ai

• Fan out, don’t queue. When one source feeds many deliverables, one agent per deliverable with shared voice rules keeps the pieces consistent.
• One profile file holds all the show-specific facts, so the same skill works for any show.
• Flag, don’t guess. Uncertain names and risky claims get surfaced to a human.

Caveats

• This is a first-person post by the tool’s author about her own tool. It is free and under an MIT license. The post does not sell a product, course or paid tier.
• A human stays in the loop. The skill refuses to do the Riverside export, publishing or opening a pull request without asking first. It flags uncertain guest and brand spellings and legal-risk claims instead of guessing, and the post says it never invents a quote. She writes: “You still make the calls on the final title and anything with reputation weight.”
• The edit phase only works with browser-based editors like Riverside, and it needs Chrome with the Claude in Chrome tools.
• The post says eleven pieces get drafted at once, but it does not list which eleven.
• X: Unverified not checked.

Sources

• https://www.gigikenneth.com/post/riverside-podcast-content-pack (primary: Gigi’s first-person description of pod-pack)

---

Want agents like these working your distribution? See the Agent Stack Teardown. Get new teardowns weekly: subscribe (double opt-in).