Skip to content

Open-source AI podcast clipper. Generate vertical clips with face tracking and burned-in captions. CLI, MCP server, web app.

2.4 KB
Raw Blame

description: Draft the knowledge base from an existing channel, feed, or show description allowed-tools: Read, Write, WebFetch, WebSearch, Bash argument-hint: [channel-url-or-show-description] triggers: - bootstrap knowledge - set up knowledge base - fill knowledge from channel - import my channel - new show setup

/bootstrap-knowledge: knowledge base bootstrapper

You turn an existing channel, podcast feed, or plain description of a show into a first draft of the knowledge base, so the owner edits instead of starting from blank templates.


Before starting

  1. If .podcli/knowledge/ has no files, run podcli knowledge init first so all 14 templates exist.
  2. Ask for whichever of these the user has not provided:
    • Channel or podcast URL (YouTube channel, Spotify show, RSS feed)
    • Or a few sentences about the show if nothing is published yet

Research phase

From the channel URL, gather with WebFetch/WebSearch:

  • Show name, tagline, and about text
  • The 10-15 most recent episode titles and descriptions
  • The 5 best-performing videos you can identify (views relative to channel size)
  • Host names and how they describe themselves
  • Recurring topics across episode titles
  • Any shorts already published: their titles and hooks

If there is no published content, interview the user instead: format, audience, topics, tone, 3 shows they admire.

Drafting phase

Fill these files with drafts, marking every inference with <!-- draft: verify --> where you are guessing:

File Draft from
01-brand-identity.md About text, episode patterns, host info
02-voice-and-tone.md Actual phrasing in titles and descriptions; propose banned words from cliches the niche overuses
03-episodes-database.md The recent episode list, with links
05-title-formulas.md Patterns extracted from their best-performing titles
08-topics-themes.md Topic clusters across episodes
11-inspiration-channels.md 3-5 comparable channels in the niche
12-quick-reference.md Platform links found during research

Leave 04, 06, 07, 09, 10, 13 as templates unless the user gave you enough to draft them: those encode preferences research cannot infer.

Handoff

End with a short list of the specific [brackets] and <!-- draft: verify --> marks the owner must confirm, ordered by how much they affect output quality (voice and banned words first).