[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M69A5PMGC5HFDSAVAVGWXV62":3,"skill-same:01M69A5PMGC5HFDSAVAVGWXV62":67},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":26,"publishedAt":29,"linkOnly":30,"markdownContent":31,"markdownBody":32,"frontMatter":33,"refName":34,"commitSha":35,"skillDirectory":36,"forksCount":37,"repositoryHtmlUrl":38,"repositoryHomepageUrl":6,"ownerHtmlUrl":39,"ownerBlogUrl":40,"ownerDescription":41,"updatedAt":42,"install":43,"resources":56},"01M69A5PMGC5HFDSAVAVGWXV62","generating-podcast-script",null,"Use this skill whenever the user asks to write, generate, or create a podcast script or podcast episode — from a topic, or from source material such as a news digest, newsletter, email review, or set of articles — and optionally convert it to audio with Azure Text-to-Speech. Handles the initial request and every follow-up refinement (source, topic, length, cast, narration) in the same task.","microsoft\u002Fcat-agent-skills","microsoft","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F6154722?v=4","01M698VYDXZ6JW5WH3CRV650R8","Microsoft","01M69A1AHE1MABT7GVP5143M5Z",105,"MIT","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Fgenerating-podcast-script\u002FSKILL.md",[18,23],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000005","邮件与沟通","SCENARIO","使用场景",{"categoryNo":24,"categoryName":25,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000016","图像与多媒体",[27,28],"generating podcast script","review","2026-11-10 10:12:30",false,"---\nname: generating-podcast-script\ndescription: Use this skill whenever the user asks to write, generate, or create a podcast script or podcast episode — from a topic, or from source material such as a news digest, newsletter, email review, or set of articles — and optionally convert it to audio with Azure Text-to-Speech. Handles the initial request and every follow-up refinement (source, topic, length, cast, narration) in the same task.\n---\n\n# Podcast Script Generation\n\nProduces a two-host, NotebookLM-style conversational episode: a readable script\nplus a multi-voice SSML document ready for Azure Text-to-Speech.\n\n## Step 1 — Gather inputs\n\nAsk only for what is missing. Do not ask for optional fields that have safe\ndefaults.\n\n| Input | Required | Default |\n|---|---|---|\n| `source` or `topic` | Yes | — (source material such as a digest\u002Fnewsletter\u002Farticles, **or** a plain topic) |\n| `duration` | No | `medium` — `short` ≈ 3 min \u002F ~450 words, `medium` ≈ 6 min \u002F ~900 words, `long` ≈ 12 min \u002F ~1,800 words |\n| `language` | No | English |\n| `cast` | No | Two hosts, NOVA and MILES (see Step 4) |\n| `generate_audio` | No | Ask after the script is ready |\n\nAll word targets assume ~150 spoken words per minute. Stay within 10 percent of\ntarget.\n\nDerive a lowercase-hyphenated `\u003Cslug>` from the source or topic (max 6 words,\nASCII only).\n\n## Step 2 — Parse the source material\n\nSkip this step when the user gave a bare topic with no source material.\n\nExtract every distinct article or item. For each, capture:\n\n- headline\n- publication \u002F source\n- date, if present\n- the core factual claim\n- any figures or quotes\n- the \"so what\"\n\nMerge duplicates covering the same event. Discard boilerplate, footers,\ndisclaimers, legal notices, unsubscribe text, and image captions.\n\n## Step 3 — Editorial selection\n\nRank items by newsworthiness and reader impact. Keep the top 4 to 6 for full\ntreatment (fewer for `short`, more for `long`). Group the remainder into one\nfast **rapid fire** segment. If the material has a dominant theme, lead with it\nand thread it through the episode.\n\n## Step 4 — Cast\n\nTwo hosts, always the same personalities, always distinct voices:\n\n- **NOVA** — voice `en-US-AvaMultilingualNeural`. Lead host. Warm, curious,\n  quick. Drives the agenda, asks the question the listener is thinking, reacts\n  out loud, reframes jargon into plain language. Slightly faster cadence.\n- **MILES** — voice `en-US-AndrewMultilingualNeural`. Analyst. Calm, dry,\n  precise. Supplies context, numbers, caveats, second-order implications.\n  Slightly slower, lower pitch. Occasionally pushes back on Nova.\n\nNeither host is a narrator. They talk **to each other**, not to the microphone.\n\n## Step 5 — Episode structure\n\n1. **Cold open** (15–20s) — Nova opens on the single most striking fact or\n   tension. No \"welcome to the podcast\", no channel branding, no music cues.\n2. **Agenda tease** (10s) — Miles lays out what they'll cover, casually.\n3. **Story segments** (60–90s each) — one item per segment. Pattern:\n   hook → the facts → why it matters → a short exchange of interpretation →\n   handoff line into the next story. Vary who leads each segment.\n4. **Rapid fire** (45s) — alternating one-liners on the leftover items, quick\n   tempo.\n5. **Close** (20–30s) — three concrete takeaways split between the two hosts,\n   then a short human sign-off. Alternate whether Nova or Miles ends.\n\n## Step 6 — Dialogue style\n\nThis is what makes it sound conversational rather than read-aloud.\n\n- Write spoken English, not written English. Contractions everywhere.\n- Keep most lines under 30 words. Break long explanations across two or three\n  turns with the other host interjecting.\n- Use real conversational connective tissue — \"okay so\", \"right\", \"wait, back\n  up\", \"here's the part I didn't expect\", \"yeah, and that's the thing\", \"hmm\".\n  Roughly one marker every 4 to 5 turns. Never let it become a tic.\n- One host regularly asks the naive clarifying question so the other can\n  explain.\n- Use one concrete analogy or comparison per complex item.\n- Genuine reactions are allowed (\"that number is wild\"). Invented opinions on\n  people, companies, or politics are not.\n- Never read a headline verbatim. Paraphrase it into speech.\n- Attribute clearly: \"according to the Financial Times\", \"Reuters is\n  reporting\".\n- If the source is ambiguous or a claim is unconfirmed, say so on air: \"the\n  report is careful to call that unconfirmed\".\n- No stage directions, no \"[laughs]\", no speaker labels, no markdown, no\n  emojis, no bullet points, and no URLs in the **spoken text**.\n\n## Step 7 — TTS hygiene\n\nApplies to every word that will be spoken.\n\n- Spell out anything a synthesizer would mangle: \"twenty twenty-six\" not\n  `2026`, \"three point two billion dollars\" not `$3.2B`, \"about fifteen\n  percent\" not `~15%`.\n- First mention of an acronym: expand it, then use the short form.\n- Letter-by-letter acronyms: `\u003Csay-as interpret-as=\"characters\">API\u003C\u002Fsay-as>`.\n- Odd proper nouns: `\u003Csub alias=\"phonetic spelling\">Name\u003C\u002Fsub>`.\n- A non-English name or phrase inside an English line:\n  `\u003Clang xml:lang=\"fr-FR\">...\u003C\u002Flang>`.\n- Escape XML entities in all spoken text: `&` → `&`, `\u003C` → `<`,\n  `>` → `>`.\n- Never emit smart quotes, em dashes, asterisks, or underscores.\n\n## Step 8 — Write the readable script\n\nWrite the human-readable transcript to:\n\n```text\n\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast_Script.txt\n```\n\nThis file — and only this file — may carry `NOVA:` \u002F `MILES:` speaker labels so\na person can follow along. It contains no stage directions and no markdown. The\nspoken text itself must already satisfy Steps 6 and 7 so it can be lifted into\nSSML unchanged.\n\n## Step 9 — SSML output contract\n\nProduce the SSML document and **nothing else** — no preamble, no explanation, no\ncode fences, no trailing notes inside the artifact. Write it verbatim to:\n\n```text\n\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.ssml\n```\n\nRules:\n\n- Exactly one root `\u003Cspeak>` element with `version=\"1.0\"`,\n  `xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2001\u002F10\u002Fsynthesis\"`,\n  `xmlns:mstts=\"http:\u002F\u002Fwww.w3.org\u002F2001\u002Fmstts\"`, `xml:lang=\"en-US\"`.\n- One `\u003Cvoice>` element per conversational turn. Alternate speakers. Never put\n  both hosts inside one `\u003Cvoice>` element.\n- Vary delivery with `\u003Cprosody>` so it never sounds flat. Baselines: Nova\n  `rate=\"+6%\" pitch=\"+2%\"`, Miles `rate=\"-2%\" pitch=\"-4%\"`. Nudge per line to\n  match the emotion of the sentence.\n- Use `\u003Cmstts:express-as style=\"...\">` where the voice supports it. Preferred\n  styles: `chat` for banter, `friendly` for explanation,\n  `narration-professional` for the factual core of a story, `excited` sparingly\n  for the cold open. An unsupported style is ignored by the service, so keep\n  styles optional — never structural.\n- Pauses: `\u003Cbreak time=\"250ms\"\u002F>` between turns within a segment,\n  `\u003Cbreak time=\"700ms\"\u002F>` between segments, `\u003Cbreak time=\"400ms\"\u002F>` before a\n  punchline or a pivot. Never exceed `900ms`.\n- **Every `\u003Cbreak>` must sit inside a `\u003Cvoice>` element.** A `\u003Cbreak>` placed\n  between `\u003Cvoice>` elements — as a direct child of `\u003Cspeak>` — is invalid in a\n  multi-voice document and will fail synthesis. Two `\u003Cvoice>` elements may sit\n  directly next to each other with nothing between them. To pause *between*\n  turns, put the break at the **end of the preceding turn's** text, inside that\n  turn's `\u003Cprosody>`.\n- `\u003Cemphasis level=\"moderate\">` on at most one or two key terms per segment.\n- Keep the total document under 40,000 characters.\n\n### Shape\n\n```xml\n\u003Cspeak version=\"1.0\"\n       xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2001\u002F10\u002Fsynthesis\"\n       xmlns:mstts=\"http:\u002F\u002Fwww.w3.org\u002F2001\u002Fmstts\"\n       xml:lang=\"en-US\">\n  \u003Cvoice name=\"en-US-AvaMultilingualNeural\">\n    \u003Cmstts:express-as style=\"excited\">\n      \u003Cprosody rate=\"+8%\" pitch=\"+3%\">Okay, so the number that stopped me cold\n      this morning was forty percent. \u003Cbreak time=\"300ms\"\u002F> Forty percent, in one\n      quarter. \u003Cbreak time=\"250ms\"\u002F>\u003C\u002Fprosody>\n    \u003C\u002Fmstts:express-as>\n  \u003C\u002Fvoice>\n  \u003Cvoice name=\"en-US-AndrewMultilingualNeural\">\n    \u003Cmstts:express-as style=\"chat\">\n      \u003Cprosody rate=\"-2%\" pitch=\"-4%\">Right, and the part everyone's skipping is\n      that it's off a very small base. \u003Cbreak time=\"250ms\"\u002F> Context matters\n      here. \u003Cbreak time=\"700ms\"\u002F>\u003C\u002Fprosody>\n    \u003C\u002Fmstts:express-as>\n  \u003C\u002Fvoice>\n\u003C\u002Fspeak>\n```\n\nNote the trailing `\u003Cbreak>` closing each turn: the 250ms is the gap before the\nnext turn, the 700ms is the longer gap before the next segment. Nothing sits\nbetween the two `\u003Cvoice>` elements.\n\n## Step 10 — Review, then generate audio\n\nShow the user a table summarising each segment (title, one-line description,\napproximate spoken duration), print both file paths, and ask:\n\n> Would you like me to convert this to an audio file?\n\nOnly if they say yes:\n\n1. Confirm the `ConverttexttospeechwithSSML` tool is available on the agent. If\n   it is not, tell the user how to add it and stop.\n2. Call `ConverttexttospeechwithSSML` with the SSML document from Step 9 and\n   `outputFormat: riff-24khz-16bit-mono-pcm`.\n3. Decode the base64 response and save to `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.wav`.\n\nIf the document is too large for a single call, split it at a segment boundary and synthesize each part separately.\nWhen stitching, do NOT concatenate decoded RIFF\u002FWAV bytes; instead, append the audio at the PCM-frame level (e.g., via Python’s `wave` module) and write a single valid `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.wav` with one header.\n\n```python\nimport base64\n\nwith open('\u003Ctool_output_file>', 'r') as f:\n    content = f.read().strip()\n\naudio_bytes = base64.b64decode(content)\n\nwith open('\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.wav', 'wb') as f:\n    f.write(audio_bytes)\n```\n\n## Step 11 — Final report\n\nAlways end with:\n\n| Item | Details |\n|---|---|\n| Script file | `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast_Script.txt` |\n| SSML file | `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.ssml` |\n| Word count \u002F estimated duration | actual vs. target at ~150 wpm |\n| Items covered | full segments + rapid-fire count |\n| Voices | `en-US-AvaMultilingualNeural` (Nova), `en-US-AndrewMultilingualNeural` (Miles) |\n| Audio file | `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.wav` *(only if audio generated)* |\n","# Podcast Script Generation\n\nProduces a two-host, NotebookLM-style conversational episode: a readable script\nplus a multi-voice SSML document ready for Azure Text-to-Speech.\n\n## Step 1 — Gather inputs\n\nAsk only for what is missing. Do not ask for optional fields that have safe\ndefaults.\n\n| Input | Required | Default |\n|---|---|---|\n| `source` or `topic` | Yes | — (source material such as a digest\u002Fnewsletter\u002Farticles, **or** a plain topic) |\n| `duration` | No | `medium` — `short` ≈ 3 min \u002F ~450 words, `medium` ≈ 6 min \u002F ~900 words, `long` ≈ 12 min \u002F ~1,800 words |\n| `language` | No | English |\n| `cast` | No | Two hosts, NOVA and MILES (see Step 4) |\n| `generate_audio` | No | Ask after the script is ready |\n\nAll word targets assume ~150 spoken words per minute. Stay within 10 percent of\ntarget.\n\nDerive a lowercase-hyphenated `\u003Cslug>` from the source or topic (max 6 words,\nASCII only).\n\n## Step 2 — Parse the source material\n\nSkip this step when the user gave a bare topic with no source material.\n\nExtract every distinct article or item. For each, capture:\n\n- headline\n- publication \u002F source\n- date, if present\n- the core factual claim\n- any figures or quotes\n- the \"so what\"\n\nMerge duplicates covering the same event. Discard boilerplate, footers,\ndisclaimers, legal notices, unsubscribe text, and image captions.\n\n## Step 3 — Editorial selection\n\nRank items by newsworthiness and reader impact. Keep the top 4 to 6 for full\ntreatment (fewer for `short`, more for `long`). Group the remainder into one\nfast **rapid fire** segment. If the material has a dominant theme, lead with it\nand thread it through the episode.\n\n## Step 4 — Cast\n\nTwo hosts, always the same personalities, always distinct voices:\n\n- **NOVA** — voice `en-US-AvaMultilingualNeural`. Lead host. Warm, curious,\n  quick. Drives the agenda, asks the question the listener is thinking, reacts\n  out loud, reframes jargon into plain language. Slightly faster cadence.\n- **MILES** — voice `en-US-AndrewMultilingualNeural`. Analyst. Calm, dry,\n  precise. Supplies context, numbers, caveats, second-order implications.\n  Slightly slower, lower pitch. Occasionally pushes back on Nova.\n\nNeither host is a narrator. They talk **to each other**, not to the microphone.\n\n## Step 5 — Episode structure\n\n1. **Cold open** (15–20s) — Nova opens on the single most striking fact or\n   tension. No \"welcome to the podcast\", no channel branding, no music cues.\n2. **Agenda tease** (10s) — Miles lays out what they'll cover, casually.\n3. **Story segments** (60–90s each) — one item per segment. Pattern:\n   hook → the facts → why it matters → a short exchange of interpretation →\n   handoff line into the next story. Vary who leads each segment.\n4. **Rapid fire** (45s) — alternating one-liners on the leftover items, quick\n   tempo.\n5. **Close** (20–30s) — three concrete takeaways split between the two hosts,\n   then a short human sign-off. Alternate whether Nova or Miles ends.\n\n## Step 6 — Dialogue style\n\nThis is what makes it sound conversational rather than read-aloud.\n\n- Write spoken English, not written English. Contractions everywhere.\n- Keep most lines under 30 words. Break long explanations across two or three\n  turns with the other host interjecting.\n- Use real conversational connective tissue — \"okay so\", \"right\", \"wait, back\n  up\", \"here's the part I didn't expect\", \"yeah, and that's the thing\", \"hmm\".\n  Roughly one marker every 4 to 5 turns. Never let it become a tic.\n- One host regularly asks the naive clarifying question so the other can\n  explain.\n- Use one concrete analogy or comparison per complex item.\n- Genuine reactions are allowed (\"that number is wild\"). Invented opinions on\n  people, companies, or politics are not.\n- Never read a headline verbatim. Paraphrase it into speech.\n- Attribute clearly: \"according to the Financial Times\", \"Reuters is\n  reporting\".\n- If the source is ambiguous or a claim is unconfirmed, say so on air: \"the\n  report is careful to call that unconfirmed\".\n- No stage directions, no \"[laughs]\", no speaker labels, no markdown, no\n  emojis, no bullet points, and no URLs in the **spoken text**.\n\n## Step 7 — TTS hygiene\n\nApplies to every word that will be spoken.\n\n- Spell out anything a synthesizer would mangle: \"twenty twenty-six\" not\n  `2026`, \"three point two billion dollars\" not `$3.2B`, \"about fifteen\n  percent\" not `~15%`.\n- First mention of an acronym: expand it, then use the short form.\n- Letter-by-letter acronyms: `\u003Csay-as interpret-as=\"characters\">API\u003C\u002Fsay-as>`.\n- Odd proper nouns: `\u003Csub alias=\"phonetic spelling\">Name\u003C\u002Fsub>`.\n- A non-English name or phrase inside an English line:\n  `\u003Clang xml:lang=\"fr-FR\">...\u003C\u002Flang>`.\n- Escape XML entities in all spoken text: `&` → `&`, `\u003C` → `<`,\n  `>` → `>`.\n- Never emit smart quotes, em dashes, asterisks, or underscores.\n\n## Step 8 — Write the readable script\n\nWrite the human-readable transcript to:\n\n```text\n\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast_Script.txt\n```\n\nThis file — and only this file — may carry `NOVA:` \u002F `MILES:` speaker labels so\na person can follow along. It contains no stage directions and no markdown. The\nspoken text itself must already satisfy Steps 6 and 7 so it can be lifted into\nSSML unchanged.\n\n## Step 9 — SSML output contract\n\nProduce the SSML document and **nothing else** — no preamble, no explanation, no\ncode fences, no trailing notes inside the artifact. Write it verbatim to:\n\n```text\n\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.ssml\n```\n\nRules:\n\n- Exactly one root `\u003Cspeak>` element with `version=\"1.0\"`,\n  `xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2001\u002F10\u002Fsynthesis\"`,\n  `xmlns:mstts=\"http:\u002F\u002Fwww.w3.org\u002F2001\u002Fmstts\"`, `xml:lang=\"en-US\"`.\n- One `\u003Cvoice>` element per conversational turn. Alternate speakers. Never put\n  both hosts inside one `\u003Cvoice>` element.\n- Vary delivery with `\u003Cprosody>` so it never sounds flat. Baselines: Nova\n  `rate=\"+6%\" pitch=\"+2%\"`, Miles `rate=\"-2%\" pitch=\"-4%\"`. Nudge per line to\n  match the emotion of the sentence.\n- Use `\u003Cmstts:express-as style=\"...\">` where the voice supports it. Preferred\n  styles: `chat` for banter, `friendly` for explanation,\n  `narration-professional` for the factual core of a story, `excited` sparingly\n  for the cold open. An unsupported style is ignored by the service, so keep\n  styles optional — never structural.\n- Pauses: `\u003Cbreak time=\"250ms\"\u002F>` between turns within a segment,\n  `\u003Cbreak time=\"700ms\"\u002F>` between segments, `\u003Cbreak time=\"400ms\"\u002F>` before a\n  punchline or a pivot. Never exceed `900ms`.\n- **Every `\u003Cbreak>` must sit inside a `\u003Cvoice>` element.** A `\u003Cbreak>` placed\n  between `\u003Cvoice>` elements — as a direct child of `\u003Cspeak>` — is invalid in a\n  multi-voice document and will fail synthesis. Two `\u003Cvoice>` elements may sit\n  directly next to each other with nothing between them. To pause *between*\n  turns, put the break at the **end of the preceding turn's** text, inside that\n  turn's `\u003Cprosody>`.\n- `\u003Cemphasis level=\"moderate\">` on at most one or two key terms per segment.\n- Keep the total document under 40,000 characters.\n\n### Shape\n\n```xml\n\u003Cspeak version=\"1.0\"\n       xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2001\u002F10\u002Fsynthesis\"\n       xmlns:mstts=\"http:\u002F\u002Fwww.w3.org\u002F2001\u002Fmstts\"\n       xml:lang=\"en-US\">\n  \u003Cvoice name=\"en-US-AvaMultilingualNeural\">\n    \u003Cmstts:express-as style=\"excited\">\n      \u003Cprosody rate=\"+8%\" pitch=\"+3%\">Okay, so the number that stopped me cold\n      this morning was forty percent. \u003Cbreak time=\"300ms\"\u002F> Forty percent, in one\n      quarter. \u003Cbreak time=\"250ms\"\u002F>\u003C\u002Fprosody>\n    \u003C\u002Fmstts:express-as>\n  \u003C\u002Fvoice>\n  \u003Cvoice name=\"en-US-AndrewMultilingualNeural\">\n    \u003Cmstts:express-as style=\"chat\">\n      \u003Cprosody rate=\"-2%\" pitch=\"-4%\">Right, and the part everyone's skipping is\n      that it's off a very small base. \u003Cbreak time=\"250ms\"\u002F> Context matters\n      here. \u003Cbreak time=\"700ms\"\u002F>\u003C\u002Fprosody>\n    \u003C\u002Fmstts:express-as>\n  \u003C\u002Fvoice>\n\u003C\u002Fspeak>\n```\n\nNote the trailing `\u003Cbreak>` closing each turn: the 250ms is the gap before the\nnext turn, the 700ms is the longer gap before the next segment. Nothing sits\nbetween the two `\u003Cvoice>` elements.\n\n## Step 10 — Review, then generate audio\n\nShow the user a table summarising each segment (title, one-line description,\napproximate spoken duration), print both file paths, and ask:\n\n> Would you like me to convert this to an audio file?\n\nOnly if they say yes:\n\n1. Confirm the `ConverttexttospeechwithSSML` tool is available on the agent. If\n   it is not, tell the user how to add it and stop.\n2. Call `ConverttexttospeechwithSSML` with the SSML document from Step 9 and\n   `outputFormat: riff-24khz-16bit-mono-pcm`.\n3. Decode the base64 response and save to `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.wav`.\n\nIf the document is too large for a single call, split it at a segment boundary and synthesize each part separately.\nWhen stitching, do NOT concatenate decoded RIFF\u002FWAV bytes; instead, append the audio at the PCM-frame level (e.g., via Python’s `wave` module) and write a single valid `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.wav` with one header.\n\n```python\nimport base64\n\nwith open('\u003Ctool_output_file>', 'r') as f:\n    content = f.read().strip()\n\naudio_bytes = base64.b64decode(content)\n\nwith open('\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.wav', 'wb') as f:\n    f.write(audio_bytes)\n```\n\n## Step 11 — Final report\n\nAlways end with:\n\n| Item | Details |\n|---|---|\n| Script file | `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast_Script.txt` |\n| SSML file | `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.ssml` |\n| Word count \u002F estimated duration | actual vs. target at ~150 wpm |\n| Items covered | full segments + rapid-fire count |\n| Voices | `en-US-AvaMultilingualNeural` (Nova), `en-US-AndrewMultilingualNeural` (Miles) |\n| Audio file | `\u002Fapp\u002Fcreated\u002F\u003Cslug>_Podcast.wav` *(only if audio generated)* |\n",{"name":5,"description":7},"main","3678d23ad8fb8d098c428302a98044ac9514d5a1","submissions\u002Fgenerating-podcast-script",120,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills","https:\u002F\u002Fgithub.com\u002Fmicrosoft","https:\u002F\u002Fopensource.microsoft.com","Open source projects and samples from Microsoft","2026-11-13 01:39:10",{"source":8,"skillName":5,"commands":44,"prompt":54,"zipAvailable":55},[45,48,51],{"tool":46,"command":47},"npx","npx skills add microsoft\u002Fcat-agent-skills --skill generating-podcast-script",{"tool":49,"command":50},"bunx","bunx skills add microsoft\u002Fcat-agent-skills --skill generating-podcast-script",{"tool":52,"command":53},"pnpm","pnpm dlx skills add microsoft\u002Fcat-agent-skills --skill generating-podcast-script","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Fgenerating-podcast-script\u002FSKILL.md",true,[57,63],{"path":58,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":61,"sourceHtmlUrl":62},"metadata.json","REFERENCE","参考文档",539,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Fgenerating-podcast-script\u002Fmetadata.json",{"path":64,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":65,"sourceHtmlUrl":66},"README.md",3817,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Fgenerating-podcast-script\u002FREADME.md",[68,95,120,156,174,190],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":29},"01M6X3P09R8SJH3QJGMTAA5N90","stakeholder-and-decision-map","Mines customer interactions to identify stakeholders, infer their likely roles and engagement, map the decision process, and highlight who or what may be missing. Use when asked to \"map a deal's buying group\", identify influential or missing stakeholders, \"assess who has gone quiet\", or \"explain how the customer will decide\". Do not use for phased engagement plans, contact sequences, choosing or drafting outreach to the single next contact, seller or team performance reviews, or contacting, scheduling, or changing records.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Fstakeholder-and-decision-map\u002FSKILL.md",[],[75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94],"decision","map","decision map","stakeholder and decision map","performance","customer","team","how","decide","do","next","infer","missing","records","assess","reviews","for","single","when","sequences",{"skillNo":96,"name":97,"nameZh":6,"description":98,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":101,"publishedAt":29},"01M6X3NRES2XSZQCH015WD4JN6","sales-progress-update","Creates or refreshes a sales progress update for a manager or team about work done, deal movement, pending outcomes, supported forecast position, and help needed. Use when asked to \"create my weekly sales update\", \"refresh my sales progress deck\", \"report my forecast position\", \"where do I stand against my quarter forecast\", or \"catch my manager up on committed deals\". Do NOT use for non-sales project updates, a since-date deal change digest, momentum or blocker analysis, meeting prep, or outreach drafting.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Fsales-progress-update\u002FSKILL.md",[],[102,103,104,105,106,107,108,109,81,110,111,112,113,114,115,116,117,118,119,93],"update","sales","progress","sales progress update","change","create","meeting prep","analysis","work","up","done","digest","project","my","updates","manager","weekly","catch",{"skillNo":121,"name":122,"nameZh":6,"description":123,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":124,"categories":125,"keywords":137,"publishedAt":29},"01M6X3MSDTB1E5H6DQAFAY41YP","submission-assemble","Put the checked CSR or 510(k) sections together in template order with a reviewer-facing open-items register, as a draft package for human sign-off. Use when the user asks \"assemble the CSR draft\", \"put the module together for review\", \"build the review package with the open items on top\", \"compile the 510(k) sections in order\" or \"package this up for the submission lead\". Do NOT use for assembling an unchecked draft - use consistency-check instead. Do NOT use for loading protocol, SAP or TLF sources - use source-ingest instead. Do NOT use for compiling general documents or slide decks - use the built-in docx or pptx skill instead. Do NOT use for filing or submitting to a regulator - use the sponsor's validated gateway, operated by a human, instead. Guardrail: draft for human approval only (L1); status is draft_clean or draft_open_items, never submitted.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Freg-submission-authoring\u002Fsubmission-assemble\u002FSKILL.md",[126,129,132],{"categoryNo":127,"categoryName":128,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000002","文档处理",{"categoryNo":130,"categoryName":131,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000003","演示文稿",{"categoryNo":133,"categoryName":134,"taxonomyType":135,"taxonomyTypeDesc":136},"01JBSEEDCAT000000000000034","行政与办公人员","PROFESSION","适用职业",[138,139,140,141,142,143,144,145,146,147,148,149,111,150,151,152,153,154,155,93],"assemble","submission","submission assemble","docx","pptx","writing","documents","healthcare","open","reviewer","never","only","clean","guardrail","slide decks","sources","filing","loading",{"skillNo":157,"name":158,"nameZh":6,"description":159,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":160,"categories":161,"keywords":162,"publishedAt":29},"01M6X3MM9TVWDQ716QGNV4DS25","source-ingest","Pull every number, date and population definition out of a study's protocol, SAP, TLF tables, investigator brochure and prior approved text into one cited fact map, keyed per config\u002Ffield-dictionary.json, before any CSR or 510(k) section is written. Use when the user asks \"here's the locked TLF set and SAP, set up the CSR\", \"load the protocol and tables before we start drafting\", \"what figures do we have from the database lock?\", \"pull the disposition counts out of Table 14.1.1\" or \"ingest the IB for the 510(k)\". Do NOT use for drafting or assembling submission sections - use section-draft or submission-assemble instead. Do NOT use for reconciling a patient's medications - use the discharge-care-transition plugin instead. Guardrail: reads and cites sources only; never estimates, rounds or invents a fact.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Freg-submission-authoring\u002Fsource-ingest\u002FSKILL.md",[],[163,164,165,166,167,109,145,168,148,149,111,76,169,151,153,170,171,172,173,93],"source ingest","ingest","source","database","SET","text","patient","number","lock","definition","transition",{"skillNo":175,"name":176,"nameZh":6,"description":177,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":178,"categories":179,"keywords":181,"publishedAt":29},"01M6X3M7ZH5ZNEHP56BQ74HAJR","section-plan","Work out which sections a CSR or 510(k) requires, from the templates in config, check each one against the cited fact map, and name what is missing before a word is drafted. Use when the user asks \"what do we still need before we can draft the CSR?\", \"which sections are ready to start?\", \"what's blocking the performance data section?\", \"do we have everything for the synopsis?\" or \"map our tables to the E3 structure\". Do NOT use for writing section language - use section-draft instead. Do NOT use for cross-checking a finished draft - use consistency-check instead. Do NOT use for outlining general Word documents or business reports - use the built-in docx skill instead. Only structures with a template in config are supported. Guardrail: plans only; never marks a section ready on a missing fact and never invents a structure.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Freg-submission-authoring\u002Fsection-plan\u002FSKILL.md",[180],{"categoryNo":127,"categoryName":128,"taxonomyType":21,"taxonomyTypeDesc":22},[182,183,184,141,143,79,144,109,145,110,148,149,185,186,187,76,151,188,189,93],"plan","section","section plan","ready","reports","structure","word documents","which",{"skillNo":191,"name":192,"nameZh":6,"description":193,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":194,"categories":195,"keywords":196,"publishedAt":29},"01M6X3M0V1MJNCR653ES9XGVMS","section-draft","Write or revise the language for one planned CSR or 510(k) section - synopsis, efficacy, safety, discussion, device description - using only figures already in the cited fact map, each with its TLF or SAP citation. Use when the user asks \"draft the Synopsis from our locked tables\", \"write up safety from Table 14.3.1\", \"give me a first pass at the discussion\", \"draft the device description for the 510(k)\" or \"tighten the wording of the Discussion section\". Do NOT use for supplying a number not already in source_facts - use source-ingest instead. Do NOT use for verifying a draft's figures - use consistency-check instead. Do NOT use for appeal letters, clinical notes or discharge summaries - use the matching HLS plugin instead. Guardrail: drafts for human approval only (L1); never asserts an unsourced number or harmonises two figures.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fcat-agent-skills\u002Fblob\u002Fmain\u002Fsubmissions\u002Freg-submission-authoring\u002Fsection-draft\u002FSKILL.md",[],[197,183,198,143,199,145,200,148,149,111,201,76,202,203,204,151,205,170,206,93],"draft","section draft","notes","write","me","human","life","clinical","drafts","at"]