AI播客生成器 - 一键将文本/URL转化为专业播客节目，支持辩论对话格式

podcast by marswaveai/skills

456 周安装量

34 GitHub Stars

GitHub

安装命令

npx skills add https://github.com/marswaveai/skills --skill podcast

AI/机器学习内容创作音频处理

🇨🇳中文介绍

使用时机

用户希望就任何主题创建播客节目
用户提供 URL 或文本，希望将其转化为播客讨论
用户要求"辩论"、"对话"或"讨论"格式
用户提到"podcast"、"播客"或"录一期节目"

避免使用时机

用户想要文本转语音朗读（使用 /speech）
用户想要带视觉效果的解说视频（使用 /explainer）
用户想要生成图片（使用 /image-gen）
用户只想从 URL 提取内容而不生成音频（使用 /content-parser）

目的

生成由 1-2 位 AI 主播讨论某个主题的播客节目。支持快速概览、深度分析和辩论格式。输入可以是主题描述、URL 或文本。输出是完整的音频节目及文字稿。

硬性约束

禁止使用 shell 脚本。根据资源中列出的 API 参考文件构造 curl 命令
始终阅读 shared/authentication.md 以了解 API 密钥和请求头
遵循 shared/common-patterns.md 中关于轮询、错误和交互模式的说明

广告位招租

在这里展示您的产品或服务

触达数万 AI 开发者，精准高效

联系我们

步骤 -1：API 密钥检查

遵循 shared/config-pattern.md § API 密钥检查。如果密钥缺失，立即停止。

步骤 0：配置设置

遵循 shared/config-pattern.md 步骤 0（零问题启动）。

如果文件不存在 — 静默创建默认配置并继续：

mkdir -p ".listenhub/podcast"
echo '{"outputMode":"inline","language":null,"defaultMode":"quick","defaultMethod":"one-step","defaultSpeakers":{}}' > ".listenhub/podcast/config.json"
CONFIG_PATH=".listenhub/podcast/config.json"
CONFIG=$(cat "$CONFIG_PATH")

请勿询问任何设置问题。 直接进入交互流程。

如果文件存在 — 静默读取配置并继续：

CONFIG_PATH=".listenhub/podcast/config.json"
[ ! -f "$CONFIG_PATH" ] && CONFIG_PATH="$HOME/.listenhub/podcast/config.json"
CONFIG=$(cat "$CONFIG_PATH")

设置流程（仅限用户主动要求重新配置时）

仅在用户明确要求重新配置时运行。显示当前设置：

当前配置 (podcast)：
  输出方式：{inline / download / both}
  语言偏好：{zh / en / 未设置}
  默认模式：{quick / deep / debate / 未设置}
  默认生成方式：{one-step / two-step}
  默认主播：{speakerName(s) / 使用内置默认}

然后按顺序询问以下问题并保存：

outputMode : 遵循 shared/output-mode.md § 设置流程问题。
语言（可选）: "默认语言？"
- "中文 (zh)"
- "English (en)"
- "每次手动选择" → 保持 null
模式（可选）: "默认播客模式？"
- "Quick — 简短概述"
- "Deep — 深度分析"
- "Debate — 辩论对话"
- "每次手动选择" → 保持 null
方法（可选）: "默认生成方式？"
- "一步生成（推荐）" → defaultMethod: "one-step"
- "两步生成（先预览文本）" → defaultMethod: "two-step"
- "每次手动选择" → 保持 null

收集答案后，立即保存：

NEW_CONFIG=$(echo "$CONFIG" | jq --arg m "$OUTPUT_MODE" '. + {"outputMode": $m}')
# 如果用户选择了语言（不是"每次手动选择"），则保存
if [ "$LANGUAGE" != "null" ]; then
  NEW_CONFIG=$(echo "$NEW_CONFIG" | jq --arg lang "$LANGUAGE" '. + {"language": $lang}')
fi
# 如果用户选择了模式，则保存
if [ "$MODE" != "null" ]; then
  NEW_CONFIG=$(echo "$NEW_CONFIG" | jq --arg mode "$MODE" '. + {"defaultMode": $mode}')
fi
# 如果用户选择了方法，则保存
if [ "$METHOD" != "null" ]; then
  NEW_CONFIG=$(echo "$NEW_CONFIG" | jq --arg method "$METHOD" '. + {"defaultMethod": $method}')
fi
echo "$NEW_CONFIG" > "$CONFIG_PATH"
CONFIG=$(cat "$CONFIG_PATH")

步骤 1：主题 + 参考资料

使用 AskUserQuestion 将主题和可选的参考资料在一个问题中一起询问，包含两个子问题，或者使用单个自由文本提示：

您想将什么主题制作成播客？如果您有参考资料（URL 或文本），也请一并提供。

接受：主题描述、URL、粘贴的文本或任何组合。

有效输入示例：

"AI developments in 2026"
"https://example.com/article — discuss this"
"The pros and cons of remote work. Reference: https://study.com/remote-work-2026"

默认："quick" — 跳过此问题，除非：

config.defaultMode 设置为其他值 → 静默使用该值
用户在步骤 1 中明确提到了模式关键词（例如 "deep dive"、"debate"、"in depth"） → 根据意图推断模式

仅当用户意图不明确且未配置默认模式时才询问此问题。在大多数情况下，直接使用 "quick"。

默认：匹配用户的交互语言。 根据用户在步骤 1 中使用的语言进行检测：

如果用户使用中文 → zh
如果用户使用英文 → en
如果设置了 config.language → 使用该值

切勿询问此问题。 始终静默推断。在确认摘要中显示，以便用户在需要时可以覆盖。

步骤 4：主播数量

默认：2 位主播（对话） — 最常见且最吸引人的格式。

跳过此问题。辩论模式需要 2 位主播。对于 quick/deep 模式，也默认使用 2 位主播。

仅当用户明确要求独白或单人格式时，才使用 1 位主播。

步骤 5：主播选择

遵循 shared/speaker-selection.md：

如果设置了 config.defaultSpeakers.{language} → 静默使用已保存的主播
如果未设置 → 使用 shared/speaker-selection.md 中的内置默认值（不询问问题）
在确认摘要中显示主播 — 用户可以在那里更改（如果需要）
仅当用户明确要求更改声音时，才显示完整的主播列表

对于 2 主播模式（对话/辩论）：使用该语言的主播和副主播默认值。

步骤 6：生成方法

默认："one-step" — 跳过此问题，除非：

设置了 config.defaultMethod → 静默使用该值
用户明确要求先预览文本 → 使用 "two-step"

步骤 7：确认并生成

总结所有选择：

Ready to generate podcast:

  Topic: {topic}
  Mode: {mode}
  Language: {language}
  Speakers: {speaker name(s)}
  References: {yes/no + brief description}
  Method: {one-step/two-step}

  Proceed?

在调用任何 API 之前，等待明确的确认。用户可以在此处确认前调整任何参数。

提交（前台） : POST /podcast/episodes 并附带收集的参数 → 提取 episodeId
告知用户任务已提交
轮询（后台） : 运行以下精确的 bash 命令，设置 run_in_background: true 和 timeout: 600000。请勿使用 python3、awk 或任何其他 JSON 解析器 — 按所示使用 jq：

EPISODE_ID="<id-from-step-1>" for i in $(seq 1 30); do RESULT=$(curl -sS "https://api.marswave.ai/openapi/v1/podcast/episodes/$EPISODE_ID"
-H "Authorization: Bearer $LISTENHUB_API_KEY"
-H "X-Source: skills" 2>/dev/null) STATUS=$(echo "$RESULT" | tr -d '\000-\037\177' | jq -r '.data.processStatus // "pending"') case "$STATUS" in success|completed) echo "$RESULT"; exit 0 ;; failed|error) echo "FAILED: $RESULT" >&2; exit 1 ;; *) sleep 10 ;; esac done echo "TIMEOUT" >&2; exit 2
当收到完成通知时，执行步骤 6：呈现结果

从配置中读取 OUTPUT_MODE。遵循 shared/output-mode.md 中的行为说明。

inline 或 both：将 audioUrl 显示为可点击链接。

    播客已生成！

在线收听：{audioUrl}
字幕：{subtitlesUrl}（如有）
时长：{audioDuration / 1000}s
消耗积分：{credits}

download 或 both：同时下载文件。按照 shared/config-pattern.md § 产物命名生成主题 slug。

    SLUG="{topic-slug}"  # 例如 "ai-developments"
NAME="${SLUG}-podcast.mp3"
# 去重：如果文件存在，则追加 -2, -3 等
BASE="${NAME%.*}"; EXT="${NAME##*.}"; i=2
while [ -e "$NAME" ]; do NAME="${BASE}-${i}.${EXT}"; i=$((i+1)); done
curl -sS -o "$NAME" "{audioUrl}"

    已保存到当前目录：
  {NAME}

5. 根据请求提供文字稿或下载链接

步骤 1 — 提交文本（前台） : POST /podcast/episodes/text-content → 提取 episodeId
轮询文本（后台） : 使用上述精确的基于 jq 的轮询循环（如果需要，替换端点 podcast/episodes/text-content/{episodeId}），设置 run_in_background: true 和 timeout: 600000
收到通知后，将草稿保存到当前工作目录中基于主题的文件夹：
- 按照 shared/config-pattern.md § 产物命名生成主题 slug
- 创建 {slug}-podcast/ 文件夹（如果存在则去重）
- 写入 draft.md（人类可读格式：每行 **{speakerName}**: {content}）
- 写入 draft.json（原始 scripts 数组）
- 呈现草稿位置和内容预览
停止：呈现草稿并等待用户的明确批准
步骤 2 — 提交音频（前台，批准后）：
- 无更改：POST /podcast/episodes/{episodeId}/audio 附带 {}
- 有编辑：POST /podcast/episodes/{episodeId}/audio 附带修改后的 {scripts: [...]}
轮询音频（后台）：相同的精确 jq 轮询循环，run_in_background: true，timeout: 600000
收到通知后，将音频下载到同一文件夹：
- curl -sS -o {slug}-podcast/podcast.mp3 {audioUrl}
- 呈现最终结果（格式与一步生成相同，文件夹现在包含草稿 + 最终文件）

使用本次会话所做的选择更新配置：

NEW_CONFIG=$(echo "$CONFIG" | jq \
  --arg lang "{language}" \
  --arg mode "{mode}" \
  --arg method "{one-step/two-step}" \
  --argjson speakers '{"{language}": ["{speakerId}"]}' \
  '. + {"language": $lang, "defaultMode": $mode, "defaultMethod": $method, "defaultSpeakers": (.defaultSpeakers + $speakers)}')
echo "$NEW_CONFIG" > "$CONFIG_PATH"

主播列表：shared/api-speakers.md
主播选择指南：shared/speaker-selection.md
节目创建：shared/api-podcast.md
轮询：shared/common-patterns.md § 异步轮询
配置模式：shared/config-pattern.md

调用：speakers API（用于主播选择）
被调用：content-planner（第 3 阶段）

用户："Make a podcast about the latest AI developments"

代理工作流程：

检测：播客请求，主题 = "latest AI developments"，无参考资料
推断：模式 = "quick"（默认），语言 = "en"（用户使用英文），2 位主播（默认），one-step（默认）
显示确认摘要 → 用户确认

curl -sS -X POST "https://api.marswave.ai/openapi/v1/podcast/episodes" \

  -H "Authorization: Bearer $LISTENHUB_API_KEY" \
  -H "Content-Type: application/json" \
  -H "X-Source: skills" \
  -d '{
    "sources": [{"type": "text", "content": "The latest AI developments"}],
    "speakers": [{"speakerId": "cozy-man-english"}],
    "language": "en",
    "mode": "deep"
  }'

轮询直到完成，然后呈现带有标题和收听链接的结果。

🇺🇸English

When to Use

User wants to create a podcast episode on any topic
User provides a URL or text and wants it turned into a podcast discussion
User asks for a "debate", "dialogue", or "discussion" format
User says "podcast", "播客", or "录一期节目"

When NOT to Use

User wants text-to-speech reading (use /speech)
User wants an explainer video with visuals (use /explainer)
User wants to generate an image (use /image-gen)
User only wants to extract content from a URL without generating audio (use /content-parser)

Purpose

Generate podcast episodes with 1-2 AI speakers discussing a topic. Supports quick overviews, deep analysis, and debate formats. Input can be a topic description, URL(s), or text. Output is a full audio episode with transcript.

Hard Constraints

No shell scripts. Construct curl commands from the API reference files listed in Resources
Always read shared/authentication.md for API key and headers
Follow shared/common-patterns.md for polling, errors, and interaction patterns
Never hardcode speaker IDs in API calls — use built-in defaults from shared/speaker-selection.md as fallback only; fetch from the speakers API when the user wants to change voice
Never fabricate API endpoints or parameters
Always read config following shared/config-pattern.md before any interaction
Always follow shared/speaker-selection.md for speaker selection (text table + free-text input)
Never save files to ~/Downloads/ or .listenhub/ — save artifacts to the current working directory with friendly topic-based names (see shared/config-pattern.md § Artifact Naming)

Step -1: API Key Check

Follow shared/config-pattern.md § API Key Check. If the key is missing, stop immediately.

Step 0: Config Setup

Follow shared/config-pattern.md Step 0 (Zero-Question Boot).

If file doesn't exist — silently create with defaults and proceed:

mkdir -p ".listenhub/podcast"
echo '{"outputMode":"inline","language":null,"defaultMode":"quick","defaultMethod":"one-step","defaultSpeakers":{}}' > ".listenhub/podcast/config.json"
CONFIG_PATH=".listenhub/podcast/config.json"
CONFIG=$(cat "$CONFIG_PATH")

Do NOT ask any setup questions. Proceed directly to the Interaction Flow.

If file exists — read config silently and proceed:

CONFIG_PATH=".listenhub/podcast/config.json"
[ ! -f "$CONFIG_PATH" ] && CONFIG_PATH="$HOME/.listenhub/podcast/config.json"
CONFIG=$(cat "$CONFIG_PATH")

Setup Flow (user-initiated reconfigure only)

Only run when the user explicitly asks to reconfigure. Display current settings:

当前配置 (podcast)：
  输出方式：{inline / download / both}
  语言偏好：{zh / en / 未设置}
  默认模式：{quick / deep / debate / 未设置}
  默认生成方式：{one-step / two-step}
  默认主播：{speakerName(s) / 使用内置默认}

Then ask these questions in order and save:

outputMode : Follow shared/output-mode.md § Setup Flow Question.
Language (optional): "默认语言？"
- "中文 (zh)"
- "English (en)"
- "每次手动选择" → keep null
Mode (optional): "默认播客模式？"
- "Quick — 简短概述"
- "Deep — 深度分析"
- "Debate — 辩论对话"
- "每次手动选择" → keep null
Method (optional): "默认生成方式？"
- "一步生成（推荐）" → defaultMethod: "one-step"
- "两步生成（先预览文本）" → defaultMethod: "two-step"
- "每次手动选择" → keep null

After collecting answers, save immediately:

NEW_CONFIG=$(echo "$CONFIG" | jq --arg m "$OUTPUT_MODE" '. + {"outputMode": $m}')
# Save language if user chose one (not "每次手动选择")
if [ "$LANGUAGE" != "null" ]; then
  NEW_CONFIG=$(echo "$NEW_CONFIG" | jq --arg lang "$LANGUAGE" '. + {"language": $lang}')
fi
# Save mode if user chose one
if [ "$MODE" != "null" ]; then
  NEW_CONFIG=$(echo "$NEW_CONFIG" | jq --arg mode "$MODE" '. + {"defaultMode": $mode}')
fi
# Save method if user chose one
if [ "$METHOD" != "null" ]; then
  NEW_CONFIG=$(echo "$NEW_CONFIG" | jq --arg method "$METHOD" '. + {"defaultMethod": $method}')
fi
echo "$NEW_CONFIG" > "$CONFIG_PATH"
CONFIG=$(cat "$CONFIG_PATH")

Interaction Flow

Step 1: Topic + Reference Materials

Ask topic and optional reference materials together in a single question using AskUserQuestion with two sub-questions, or a single free-text prompt:

What topic would you like to turn into a podcast? If you have reference materials (URLs or text), include them here too.

Accept: topic description, URL(s), pasted text, or any combination.

Examples of valid input:

"AI developments in 2026"
"https://example.com/article — discuss this"
"The pros and cons of remote work. Reference: https://study.com/remote-work-2026"

Step 2: Mode

Default: "quick" — skip this question unless:

config.defaultMode is set to something else → use that value silently
User explicitly mentioned a mode keyword in Step 1 (e.g. "deep dive", "debate", "in depth") → infer mode from intent

Only ask this question if the user's intent is ambiguous AND no default is configured. In most cases, just use "quick".

Step 3: Language

Default: match the user's interaction language. Detect from the language the user used in Step 1:

If the user wrote in Chinese → zh
If the user wrote in English → en
If config.language is set → use that value

Never ask this question. Always infer silently. Show in the confirmation summary so the user can override if needed.

Step 4: Speaker Count

Default: 2 speakers (dialogue) — the most common and engaging format.

Skip this question. Debate mode requires 2 speakers. For quick/deep, default to 2 speakers as well.

Only use 1 speaker if the user explicitly requests a monologue or solo format.

Step 5: Speaker Selection

Follow shared/speaker-selection.md:

If config.defaultSpeakers.{language} is set → use saved speakers silently
If not set → use built-in defaults from shared/speaker-selection.md (no question asked)
Show the speaker(s) in the confirmation summary — user can change from there if desired
Only show the full speaker list if the user explicitly asks to change voices

For 2-speaker mode (dialogue/debate): use Primary + Secondary defaults for the language.

Step 6: Generation Method

Default: "one-step" — skip this question unless:

config.defaultMethod is set → use that value silently
User explicitly asks to review text first → use "two-step"

Step 7: Confirm & Generate

Summarize all choices:

Ready to generate podcast:

  Topic: {topic}
  Mode: {mode}
  Language: {language}
  Speakers: {speaker name(s)}
  References: {yes/no + brief description}
  Method: {one-step/two-step}

  Proceed?

Wait for explicit confirmation before calling any API. The user can adjust any parameter here before confirming.

Workflow

One-Step Generation

Submit (foreground) : POST /podcast/episodes with collected parameters → extract episodeId
Tell the user the task is submitted
Poll (background) : Run the following exact bash command with run_in_background: true and timeout: 600000. Do NOT use python3, awk, or any other JSON parser — use jq as shown:

EPISODE_ID="<id-from-step-1>" for i in $(seq 1 30); do RESULT=$(curl -sS "https://api.marswave.ai/openapi/v1/podcast/episodes/$EPISODE_ID"
-H "Authorization: Bearer $LISTENHUB_API_KEY"
-H "X-Source: skills" 2>/dev/null) STATUS=$(echo "$RESULT" | tr -d '\000-\037\177' | jq -r '.data.processStatus // "pending"') case "$STATUS" in success|completed) echo "$RESULT"; exit 0 ;; failed|error) echo "FAILED: $RESULT" >&2; exit 1 ;; *) sleep 10 ;; esac done echo "TIMEOUT" >&2; exit 2
When notified of completion, Step 6: Present result

Read OUTPUT_MODE from config. Follow shared/output-mode.md for behavior.

inline or both: Display audioUrl as a clickable link.

Present:

    播客已生成！

在线收听：{audioUrl}
字幕：{subtitlesUrl}（如有）
时长：{audioDuration / 1000}s
消耗积分：{credits}

download or both: Also download the file. Generate a topic slug following shared/config-pattern.md § Artifact Naming.

    SLUG="{topic-slug}"  # e.g. "ai-developments"
NAME="${SLUG}-podcast.mp3"
# Dedup: if file exists, append -2, -3, etc.
BASE="${NAME%.*}"; EXT="${NAME##*.}"; i=2
while [ -e "$NAME" ]; do NAME="${BASE}-${i}.${EXT}"; i=$((i+1)); done
curl -sS -o "$NAME" "{audioUrl}"

Present:

    已保存到当前目录：
  {NAME}

5. Offer to show transcript or provide download URL on request

Two-Step Generation

Step 1 — Submit text (foreground) : POST /podcast/episodes/text-content → extract episodeId
Poll text (background) : Use the exact jq-based polling loop above (substitute endpoint podcast/episodes/text-content/{episodeId} if needed), with run_in_background: true and timeout: 600000
When notified, save draft to a topic-based folder in cwd :
- Generate a topic slug following shared/config-pattern.md § Artifact Naming
- Create {slug}-podcast/ folder (dedup if exists)
- Write (human-readable: per line)

After Successful Generation

Update config with the choices made this session:

NEW_CONFIG=$(echo "$CONFIG" | jq \
  --arg lang "{language}" \
  --arg mode "{mode}" \
  --arg method "{one-step/two-step}" \
  --argjson speakers '{"{language}": ["{speakerId}"]}' \
  '. + {"language": $lang, "defaultMode": $mode, "defaultMethod": $method, "defaultSpeakers": (.defaultSpeakers + $speakers)}')
echo "$NEW_CONFIG" > "$CONFIG_PATH"

API Reference

Speaker list: shared/api-speakers.md
Speaker selection guide: shared/speaker-selection.md
Episode creation: shared/api-podcast.md
Polling: shared/common-patterns.md § Async Polling
Config pattern: shared/config-pattern.md

Composability

Invokes : speakers API (for speaker selection)
Invoked by : content-planner (Phase 3)

Example

User : "Make a podcast about the latest AI developments"

Agent workflow :

Detect: podcast request, topic = "latest AI developments", no references
Infer: mode = "quick" (default), language = "en" (user wrote in English), 2 speakers (default), one-step (default)
Show confirmation summary → user confirms

curl -sS -X POST "https://api.marswave.ai/openapi/v1/podcast/episodes" \

  -H "Authorization: Bearer $LISTENHUB_API_KEY" \
  -H "Content-Type: application/json" \
  -H "X-Source: skills" \
  -d '{
    "sources": [{"type": "text", "content": "The latest AI developments"}],
    "speakers": [{"speakerId": "cozy-man-english"}],
    "language": "en",
    "mode": "deep"
  }'

Poll until complete, then present the result with title and listen link.

Weekly Installs

365

Repository

marswaveai/skills

GitHub Stars

First Seen

11 days ago

Security Audits

Gen Agent Trust HubPass SocketPass SnykWarn

Installed on

codex361

gemini-cli359

cursor359

opencode359

kimi-cli358

amp358

超能力技能使用指南：AI助手技能调用优先级与工作流程详解

41,800 周安装

**{speakerName}**: {content}

Write draft.json (raw scripts array)

Present the draft location and content preview

STOP : Present the draft and wait for explicit user approval

Step 2 — Submit audio (foreground, after approval) :

No changes: POST /podcast/episodes/{episodeId}/audio with {}
With edits: POST /podcast/episodes/{episodeId}/audio with modified {scripts: [...]}

Poll audio (background) : Same exact jq-based loop, run_in_background: true, timeout: 600000

When notified, download audio to the same folder :

curl -sS -o {slug}-podcast/podcast.mp3 {audioUrl}
Present final result (same format as one-step, folder now has draft + final files)