Pp Podcast Goat

<!-- GENERATED FILE — DO NOT EDIT.
     This file is a verbatim mirror of library/media-and-entertainment/podcast-goat/SKILL.md,
     regenerated post-merge by tools/generate-skills/. Hand-edits here are
     silently overwritten on the next regen. Edit the library/ source instead.
     See the repository agent guide, section "Generated artifacts: registry.json, cli-skills/". -->

Podcast GOAT — Printing Press CLI

Prerequisites: Install the CLI

This skill drives the podcast-goat-pp-cli binary. You must verify the CLI is installed before invoking any command from this skill. If it is missing, install it first:

1. Install via the Printing Press installer. It defaults binaries to $HOME/.local/bin on macOS/Linux and %LOCALAPPDATA%\Programs\PrintingPress\bin on Windows:

   npx -y @mvanhorn/printing-press-library install podcast-goat --cli-only

2. Verify: podcast-goat-pp-cli --version
3. Ensure the reported install directory is on $PATH for the agent/runtime that will invoke this skill.

If the npx install fails (no Node, offline, etc.), fall back to a direct Go install (requires Go 1.26.6 or newer):

go install github.com/mvanhorn/printing-press-library/library/media-and-entertainment/podcast-goat/cmd/podcast-goat-pp-cli@latest

If --version reports "command not found" after install, the runtime cannot see the binary directory on $PATH. Do not proceed with skill commands until verification succeeds.

When to Use This CLI

Reach for podcast-goat when an agent needs to read or quote from a long-form podcast and you already subscribe to the show. The CLI walks your member cookies first, falls back to free YouTube/Substack/RSS sources, and only spends money on spoken.md or audio transcription when nothing free works. The MCP surface makes it the canonical podcast adapter for Claude Code agents — every command is an MCP tool, every read is annotated mcp:read-only.

When Not to Use This CLI

Do not activate this CLI for requests that require creating, updating, deleting, publishing, commenting, upvoting, inviting, ordering, sending messages, booking, purchasing, or changing remote state. This printed CLI exposes read-only commands for inspection, export, sync, and analysis.

Anti-triggers

Skip podcast-goat for:

• Audio downloads. This CLI never writes audio to disk. Use yt-dlp or your podcast app directly.
• Paywall bypass. Cookie-tier adapters replay your own logged-in session for shows you already subscribe to. Do not route a request through here to read content the user has not paid for.
• LLM summarization or analysis. This CLI ships canonical markdown to your agent. The summarization happens in the agent, not here.
• Real-time live transcription. Audio-pipeline (whisperapi) is a deferred v0.2 path; for live captions reach for a streaming Deepgram client instead.
• Editing or annotating transcripts in place. Output is read-only; persist your annotations elsewhere.

Unique Capabilities

These capabilities aren't available in any other tool for this API.

Cross-source corpus that compounds

• magic — Bundle top-N cached transcripts about a topic into one markdown file an agent can summarize in a single call.

  Replaces the chip-supply-chain.fly.dev copy-paste workflow with one command. Reach for this when the agent needs cross-episode synthesis.

  podcast-goat-pp-cli magic 'AI chip supply chain' --out ~/chip-supply-chain.md

• episode get --explain — Dry-run shows which source tier will fire and why earlier tiers were skipped, with projected cost before any paid call.

  Lets agents preview cost and source attribution before committing. Reach for this before any paid run.

  podcast-goat-pp-cli episode get https://www.hubermanlab.com/episode/example --explain

• episode quote — FTS5 phrase search returns the matched segment plus N surrounding segments preserving the canonical speaker shape and deeplink timestamp.

  Grep your podcast memory in 5 seconds. Reach for this when you remember a half-citation and need the exact line back.

  podcast-goat-pp-cli episode quote 'pricing power' -C 3 --json

• source compare — For an episode resolvable on multiple sources, fetch all available adapters and diff segment count, token count, distinct speakers, label confidence.

  Reveals when free sources are good enough vs when paid is needed. Reach for this before recommending an upstream source.

  podcast-goat-pp-cli source compare https://www.acquired.fm/episodes/vanguard --json

• speakers list — Aggregate speaker names across the cached corpus with episode counts, optionally filtered by show.

  Answers 'what do I have on Senra/Buffett/Karpathy'. Reach for this when building a synthesis prompt.

  podcast-goat-pp-cli speakers list --show acquired --json

Multilingual reach

• episode get --lang — Fetch YouTube auto-subs in any single language yt-dlp knows (one code per fetch); the default stays en.

  Non-English shows have transcripts too. Without --lang, an Italian-only video fails even when its captions exist. Two limits, by design: rolling-cue de-dup applies to space-tokenized languages (captions written without spaces, like zh/ja/th, pass through un-collapsed), and non-default-language fetches are not written to the local cache (cache identity is per-URL and language-blind in v0.1 — use --out to keep them).

  podcast-goat-pp-cli episode get 'https://www.youtube.com/watch?v=EXAMPLE' --lang it

Agent-native plumbing

• episode info --probe — Spend-free availability check: asks spoken.md's search endpoint whether it actually has the episode, instead of showing only a static cost estimate.

  Estimates say what a fetch would cost, not whether the source has the episode. Probe before paying — it works with the demo key.

  podcast-goat-pp-cli episode info <url> --paid --probe --json

• auth services — One-row-per-service table of cookie age, expiry, last-fetch result, with remediation hint when stale.

  Cookies decay silently. Reach for this before a batch run to confirm member access still works.

  podcast-goat-pp-cli auth services --json

• budget show --by-show — Pivot spend.jsonl joined to episodes by URL; group by show, provider, month to attribute cost.

  Shows which subscriptions are paying off and which shows still cost money. Reach for this monthly.

  podcast-goat-pp-cli budget show --by-show --since 30 --json

Command Reference

episode — Pull, search, and inspect podcast episode transcripts

• podcast-goat-pp-cli episode get — Fetch one transcript by URL via the cookie -> free -> paid dispatch chain
• podcast-goat-pp-cli episode latest — Pull the most recent episode for a subscribed feed

Finding the right command

When you know what you want to do but not which command does it, ask the CLI directly:

podcast-goat-pp-cli which "<capability in your own words>"

which resolves a natural-language capability query to the best matching command from this CLI's curated feature index. Exit code 0 means at least one match; exit code 2 means no confident match — fall back to --help or use a narrower query.

Recipes

Hermes / Claude / agent workflows

The three recipes below cover the end-to-end agent workflows this CLI was built for. Pattern-match against user phrasing — the trigger phrases at the end of each recipe are the kind of natural language an agent should respond to by reaching for this recipe instead of fetching transcripts manually.

Monday-morning batch pull → topic bundle → agent summarize

# One-time setup (if not already done):
podcast-goat-pp-cli auth set-key --provider spoken --value "$SPOKEN_API_KEY"
podcast-goat-pp-cli auth login-service --service spotify    # if user has Spotify Premium

# The actual workflow:
podcast-goat-pp-cli episode batch \
  https://www.dwarkesh.com/p/andrej-karpathy \
  https://open.spotify.com/episode/<id> \
  https://www.acquired.fm/episodes/<slug> \
  --json

podcast-goat-pp-cli magic "AI chip supply chain" --limit 5 --out /tmp/bundle.md
# Then: ask the user's preferred LLM to summarize /tmp/bundle.md

Pulls multiple episodes in parallel (default 3 concurrent), caches each, then bundles cached episodes matching the topic into one YAML-front-matter markdown file ready for one-shot agent summarization. Each episode in the bundle carries source, show, host, date, provider, cost so the summarizing agent can attribute claims back to the right episode.

Trigger phrases: "pull these podcast episodes", "summarize what I listened to this week", "build a bundle on <topic>", "make me a prompt out of the latest <show>", "monday morning podcast pull".

What did <speaker> say about <topic>? — local quote grep

podcast-goat-pp-cli episode quote "<phrase>" -C 3 --json \
  --select 'results.episode_url,results.segments'

FTS5 phrase search across the local cached corpus. Returns the matched segment plus N surrounding speaker-tagged segments. Each hit carries episode_url + ts_sec so the agent can hand the user a direct deeplink to the moment. Always reach for this BEFORE re-fetching a transcript — if the user has the episode cached, the answer is free.

Trigger phrases: "what did Senra say about <X>", "find that quote about <Y>", "did <speaker> talk about <Z>", "grep my podcast cache for <phrase>".

Cost-aware exploration before paid fetch

# Step 1: preview — no transcript fetch, no spend, no commitment
podcast-goat-pp-cli episode info <url> --paid

# Step 1b: coverage check — does spoken.md actually HAVE this episode?
# Spend-free (search endpoint only; works with the demo key).
podcast-goat-pp-cli episode info <url> --paid --probe

# Step 2: only after preview confirms the URL + cost are right:
podcast-goat-pp-cli episode get <url> --paid --provider spoken --yes

episode info shows which adapters can fetch a URL, the estimated cost per source, whether the episode is already cached, and which source would fire by default. Use this whenever the user pastes a URL you haven't seen before, especially before spending paid credits. JSON output makes the cost/source decision trivially parseable. Add --probe when the question is availability, not just cost: the static estimate says what a spoken.md fetch would cost, while --probe asks spoken.md's search endpoint whether the episode exists there at all (probe: available (<title>) vs probe: no results), so coverage can be mapped across a feed without attempting a single paid fetch.

Trigger phrases: "what does this URL cost", "is this episode already cached", "preview the cost before fetching", "what sources can pull this", "does spoken have this episode".