podwise-podcast-copilot

# Podwise Podcast Copilot Use this skill to turn raw podcast, video, and audio inputs into structured outputs that are easy to read, export, or reuse. ## Goals 1. Verify that `podwise` is installed and that the API key is configured. 2. Distinguish keyword discovery from transcript-grounded answering, then choose the correct path: `search`, `list`, `popular`, `ask`, `process`, or `get`. 3. Ask for user confirmation before any `process` command because it consumes quota/credits, unless the user explicitly asked to proceed without pausing. 4. Fetch AI outputs only after `process` completes successfully and reaches `done`. 5. Return output in the shape that matches the workflow: discovery lists, transcript-grounded answers, or episode artifacts. ## Step 1: Check the Environment Run: ```bash podwise --help podwise config show ``` Treat `podwise --help` and `podwise <command> --help` as the source of truth for exact flags, subcommands, and examples. When a workflow needs command details beyond this skill, inspect help progressively: ```bash podwise search --help podwise search episode --help podwise search podcast --help podwise ask --help podwise list --help podwise process --help podwise get --help podwise get transcript --help ``` ## Step 2: Choose the Workflow ### Discover Content, Podcasts, Episodes - Use `search` when the user wants matching podcasts or episodes returned as results to browse. - Use `ask` when the user wants an answer, synthesis, or insights derived from transcript content across podcasts. - If the user wants to find episodes by keyword, or title, run `podwise search episode`. - If the user wants to find shows or feeds by podcast name or keyword, run `podwise search podcast`. - If the user asks what podcasts say about a topic, asks for takeaways, or wants transcript-grounded analysis, run `podwise ask`. - If the user wants to discover what is currently trending across Podwise, run `podwise popular`. - If the user wants to check updates from podcasts they already follow, run `podwise list episodes` or `podwise list podcasts`. ### Process an Episode - Before running any `process` command, ask whether the user wants to spend quota/credits on processing. - Skip that confirmation only when the user already made the intent explicit, for example by saying to process it now or not to pause for confirmation. - If the user provides a YouTube or Xiaoyuzhou URL and confirms processing, run `podwise process <url>`; Podwise will import it automatically. - If the user provides a local audio or video file path and confirms processing, run `podwise process <file>`; Podwise will upload it and create an episode automatically. - If the user provides a Podwise episode URL and processing may not be complete yet, ask for confirmation first, then run `podwise process <episode-url>`. ### Retrieve AI Results - If the user wants a specific artifact for an already processed episode, run `podwise get <type> <episode-url>` directly. ## Step 3: Run the Commands ### Search for Episodes or Podcasts ```bash podwise search episode "AI agent" --limit 10 --json podwise search podcast "Lex Fridman" --limit 10 --json ``` Use `--json` when the output will be parsed by another tool or step. Use `search` when the user wants a list of matching podcasts or episodes, not a synthesized answer. Use episode search when the user is looking for a specific episode, or episode title keywords. Use podcast search when the user is looking for a show by name, wants candidate podcasts to subscribe to, or is trying to identify the podcast behind an episode. Multiple words are treated as a single phrase; pass them together as one quoted query or as separate arguments. ### Discover Popular Episodes ```bash podwise popular --json ``` Use `popular` when the user wants hot, trending, or currently popular episodes rather than keyword matches. ### Check Followed Podcast Updates ```bash podwise list episodes --json podwise list episodes --json --date today podwise list episodes --json --date yesterday podwise list episodes --json --date 2026-03-01 podwise list episodes --json --latest 7 podwise list podcasts --json podwise list podcasts --json --date today podwise list podcasts --json --latest 14 ``` Use `list episodes` to see new episodes published by shows the user follows. Use `list podcasts` to see which followed podcasts have had recent updates. ### Ask Questions Across Podcast Transcripts ```bash podwise ask "the future of AI agents" ``` Use `ask` when the user wants a synthesized answer, takeaways, comparison, or insights grounded in podcast transcripts rather than a list of matching episodes. Do not use `ask` when the user only wants to locate podcasts or episodes by keyword; use `search` for that. `ask` may take up to about `60s` because Podwise searches transcripts and generates an answer. ### Process an Episode, Video, or Local File ```bash podwise process https://podwise.ai/dashboard/episodes/7360326 podwise process https://www.youtube.com/watch?v=d0-Gn_Bxf8s podwise process https://youtu.be/d0-Gn_Bxf8s podwise process https://www.xiaoyuzhoufm.com/episode/abc123 podwise process ./interview.mp3 podwise process ./meeting.wav --title "Product Review Meeting" podwise process ./demo.mp4 --title "Launch Demo Recording" --hotwords "Podwise,LLM,ASR" ``` `process` always waits for processing to finish before it exits. Because `process` consumes quota/credits, do not run it silently. Confirm with the user first unless they explicitly asked to proceed without pausing. Supported local file extensions: `.mp3 .wav .m4a .mp4 .m4v .mov .webm`. ### Retrieve AI Outputs ```bash podwise get transcript <episode-url> podwise get summary <episode-url> podwise get qa <episode-url> podwise get chapters <episode-url> podwise get mindmap <episode-url> podwise get highlights <episode-url> podwise get keywords <episode-url> ``` `podwise get` accepts only a Podwise episode URL. Do not pass a YouTube URL, a Xiaoyuzhou URL, or a local file path to `get`. For local files, follow the same pattern: run `process <file>` first, then use the resulting Podwise episode URL with `get`. ## User Request to Command Mapping - "What do podcasts say about this topic?" -> `ask` - "Summarize the main insights from podcast transcripts on this topic" -> `ask` - "Find podcast episodes about this topic" -> `search episode` - "Find a few episodes on this topic, then I will choose what to read" -> `search episode` - "Find the podcast/show named Lex Fridman" -> `search podcast` - "Show me what's trending on Podwise right now" -> `popular` - "Show today's new episodes from podcasts I follow" -> `list episodes` - "Show which followed podcasts published recently" -> `list podcasts` - "Process this YouTube video and return the transcript and summary" -> confirm first, then `process` + `get transcript` + `get summary` - "Export this episode as subtitles" -> `get transcript --format srt` or `get transcript --format vtt` - "Give me an episode recap with summary, chapters, highlights, and keywords" -> `get summary` + `get chapters` + `get highlights` + `get keywords` - "Transcribe this local audio/video file and summarize the key points" -> confirm first, then `process <file>` + `get transcript` + `get summary` ## Common Failure Cases - If `podwise` is missing or not configured correctly, stop immediately and tell the user to fix the CLI setup first. - If a local file does not exist or the extension is unsupported, stop and ask for a valid path or supported media format. - If the workflow requires `process` and the user has not agreed to spend quota/credits, pause and ask for confirmation before continuing. - If `ask` returns an error or plan-limit issue, report it directly; do not fabricate an answer. ## Output Contract For `process` and `get` workflows, include: 1. The resolved Podwise episode URL. 2. The current processing status. 3. The requested artifacts such as summary or transcript. 4. Any unavailable artifact explicitly marked as unavailable. For `ask` workflows, include: 1. The answer text. 2. Source excerpts and episode links when `--sources` is used. 3. A note if the answer is uncited because sources were not requested. For `search`, `list`, and `popular` workflows, return the result list in a scannable order and preserve the CLI's intent: discovery, updates, or trending content. For `search` workflows specifically, return matching podcasts or episodes rather than a transcript-synthesized answer.