minimal-agent 0.13.0 → 0.14.0

package/workflows/youtube-shorts.yaml

@@ -1,202 +0,0 @@
-name: youtube-shorts
-description: |
-  示例工作流：输入主题 → LLM 写 5 幕脚本 → OpenRouter 生图 ×5 → 中文 TTS ×5
-  → ffmpeg 合成竖屏 mp4。
-  这是把 4 个通用 skill（image-gen-openrouter / tts-zh / subtitle-srt /
-  video-compose）组合的样例 —— 你可以在编辑器里改 inputs / 增删节点 / 换 skill
-  去做任意视频流水线（横屏长视频、广告片、教程片都行）。
-version: "0.1"
-
-# 声明产物：new-task 启动清理残留 / completed 报实际产物 / 同任务 resume 重写不被
-# 内核「覆盖前先 Read」守卫挡（见 CLAUDE.md 任务身份语义）
-artifacts:
-  - "out/youtube-shorts/${session}/assets/script.json"
-  - "out/youtube-shorts/${session}/shorts.mp4"
-
-inputs:
-  - name: topic
-    type: string
-    required: true
-    greedy: true
-    description: "视频主题（中文，可一句话自然语言；命令行整句进 topic，不被空格拆碎）"
-  - name: openrouter_key
-    type: string
-    required: false
-    description: "【高级·用 --input openrouter_key=...】OpenRouter API key；不填则用 OPENROUTER_API_KEY 环境变量"
-  - name: aspect
-    type: enum
-    values: ["9:16", "16:9", "1:1", "4:5"]
-    default: "9:16"
-    description: "【高级·用 --input aspect=...】图片与视频宽高比"
-  - name: width
-    type: number
-    default: 1080
-    description: "【高级·用 --input width=N】视频输出宽度"
-  - name: height
-    type: number
-    default: 1920
-    description: "【高级·用 --input height=N】视频输出高度"
-
-steps:
-  # ---------- 1. 用 LLM 写 5 幕脚本（强制 5 幕以便后续 loop 用静态数组） ----------
-  - id: write_script
-    title: "1/11·写五幕脚本"
-    llm: |
-      你是短视频文案策划。请为主题"${inputs.topic}"写一个 **正好 5 幕** 的竖屏短视频。
-      每一幕包含：
-        - text：一句 8-12 字的中文旁白
-        - image_prompt：一段 30-50 词的英文图片提示词，描述这一幕的画面
-      只输出一个 JSON 对象，结构严格如下，不要包裹在 ``` 里、不要任何额外解释：
-      {"scenes":[
-        {"text":"...","image_prompt":"..."},
-        ...共 5 项...
-      ]}
-    capture:
-      text: script_json_text
-
-  # ---------- 2. 把 LLM 输出落盘为 script.json ----------
-  - id: write_script_json
-    title: "2/11·脚本落盘 script.json"
-    tool: Write
-    artifacts:
-      - "out/youtube-shorts/${session}/assets/script.json"
-    args:
-      file_path: "out/youtube-shorts/${session}/assets/script.json"
-      content: "${script_json_text}"
-
-  # ---------- 3. 解析 script.json，拆出 prompts.txt / narrations.txt（每行一个） ----------
-  - id: parse_script
-    title: "3/11·解析脚本拆分文案"
-    tool: Bash
-    args:
-      command: |
-        set -euo pipefail
-        mkdir -p out/youtube-shorts/${session}/assets out/youtube-shorts/${session}
-        jq -r '.scenes[].image_prompt' out/youtube-shorts/${session}/assets/script.json > out/youtube-shorts/${session}/assets/prompts.txt
-        jq -r '.scenes[].text' out/youtube-shorts/${session}/assets/script.json > out/youtube-shorts/${session}/assets/narrations.txt
-        # 把 narrations 转成 JSON 数组字符串，用于后续 subtitle-srt 调用
-        jq -c '[.scenes[].text]' out/youtube-shorts/${session}/assets/script.json > out/youtube-shorts/${session}/assets/narrations_json.txt
-        wc -l out/youtube-shorts/${session}/assets/prompts.txt out/youtube-shorts/${session}/assets/narrations.txt
-    capture:
-      content: parse_log
-
-  - id: assert_script_parsed
-    title: "4/11·校验脚本拆分产物"
-    type: verify
-    checks:
-      - 'file_exists:out/youtube-shorts/${session}/assets/prompts.txt'
-      - 'file_exists:out/youtube-shorts/${session}/assets/narrations.txt'
-    onFail: "脚本解析失败：out/youtube-shorts/${session}/assets/prompts.txt 或 out/youtube-shorts/${session}/assets/narrations.txt 不存在"
-
-  # ---------- 4. 生成 5 张配图（loop 静态数组 + 每轮先读对应一行 prompt） ----------
-  # ⚠️ Bash 引号 footgun：${i} 是引擎插值（shell 之前发生）。曾写 1 基算术
-  # `sed -n '$((${i}+1))p'`——shell 单引号内不做算术展开，sed 收到字面 $((0+1))p
-  # 直接报错。正解：loop 直接用 1 基序号，sed 行号即 ${i}。
-  - id: gen_images
-    title: "5/11·生成五张配图"
-    type: loop
-    over: "[1, 2, 3, 4, 5]"
-    as: i
-    do:
-      - id: read_prompt
-        title: "读第 ${i} 行画面提示词"
-        tool: Bash
-        args:
-          command: 'sed -n "${i}p" out/youtube-shorts/${session}/assets/prompts.txt'
-        capture:
-          content: prompt_text
-      - id: gen_one_image
-        title: "生成第 ${i} 张配图"
-        skill: image-gen-openrouter
-        input: 'prompt=${prompt_text} output=out/youtube-shorts/${session}/assets/img_${i}.png aspect=${inputs.aspect} key=${inputs.openrouter_key}'
-
-  # ---------- 5. 生成 5 段 TTS 音频（1 基序号，理由同 gen_images） ----------
-  - id: gen_audio
-    title: "6/11·生成五段配音"
-    type: loop
-    over: "[1, 2, 3, 4, 5]"
-    as: i
-    do:
-      - id: read_narration
-        title: "读第 ${i} 句旁白"
-        tool: Bash
-        args:
-          command: 'sed -n "${i}p" out/youtube-shorts/${session}/assets/narrations.txt'
-        capture:
-          content: narration_text
-      - id: gen_one_tts
-        title: "合成第 ${i} 段语音"
-        skill: tts-zh
-        input: 'text=${narration_text} output=out/youtube-shorts/${session}/assets/audio_${i}.mp3'
-
-  # ---------- 6. 探测各段音频时长，逗号串联 ----------
-  - id: probe_durations
-    title: "7/11·探测音频时长"
-    tool: Bash
-    args:
-      command: |
-        set -euo pipefail
-        for f in $(ls out/youtube-shorts/${session}/assets/audio_*.mp3 | sort); do
-          ffprobe -v error -show_entries format=duration -of csv=p=0 "$f"
-        done | paste -sd, -
-    capture:
-      content: durations
-
-  # ---------- 7. 合并所有 TTS 段成一条音轨 ----------
-  - id: merge_audio
-    title: "8/11·合并音轨"
-    tool: Bash
-    args:
-      command: |
-        set -euo pipefail
-        LIST=/tmp/audio_concat.txt
-        : > "$LIST"
-        for f in $(ls out/youtube-shorts/${session}/assets/audio_*.mp3 | sort); do
-          echo "file '$(pwd)/${f}'" >> "$LIST"
-        done
-        ffmpeg -y -f concat -safe 0 -i "$LIST" -c copy out/youtube-shorts/${session}/assets/audio.mp3
-        ls -la out/youtube-shorts/${session}/assets/audio.mp3
-
-  # ---------- 8. 读 narrations_json（前面 parse_script 已写好），作 SRT 文本输入 ----------
-  - id: read_narrations_json
-    title: "9/11·读取旁白清单"
-    tool: Read
-    args:
-      file_path: "out/youtube-shorts/${session}/assets/narrations_json.txt"
-    capture:
-      content: narrations_json
-
-  # ---------- 9. 生成 SRT 字幕 ----------
-  - id: make_srt
-    title: "10/11·生成字幕"
-    skill: subtitle-srt
-    input: 'texts=${narrations_json} durations=${durations} output=out/youtube-shorts/${session}/assets/subs.srt'
-
-  # ---------- 10. 合成最终 mp4 ----------
-  - id: compose_video
-    title: "11/11·合成最终视频"
-    skill: video-compose
-    input: 'images_glob=out/youtube-shorts/${session}/assets/img_*.png audio=out/youtube-shorts/${session}/assets/audio.mp3 srt=out/youtube-shorts/${session}/assets/subs.srt output=out/youtube-shorts/${session}/shorts.mp4 width=${inputs.width} height=${inputs.height}'
-
-  # ---------- 11. 守门 verify（升级自 assert） ----------
-  - id: verify_output
-    title: "完成·守门校验"
-    type: verify
-    checks:
-      - 'file_exists:out/youtube-shorts/${session}/shorts.mp4'
-    onFail: "最终视频 out/youtube-shorts/${session}/shorts.mp4 未生成，请检查 video-compose 日志"
-
-__meta:
-  layout:
-    - { id: write_script,           x: 100, y: 50 }
-    - { id: write_script_json,      x: 100, y: 200 }
-    - { id: parse_script,           x: 100, y: 350 }
-    - { id: assert_script_parsed,   x: 100, y: 500 }
-    - { id: gen_images,             x: 100, y: 650 }
-    - { id: gen_audio,              x: 450, y: 650 }
-    - { id: probe_durations,        x: 450, y: 850 }
-    - { id: merge_audio,            x: 450, y: 1000 }
-    - { id: read_narrations_json,   x: 100, y: 850 }
-    - { id: make_srt,               x: 275, y: 1150 }
-    - { id: compose_video,          x: 275, y: 1300 }
-    - { id: verify_output,          x: 275, y: 1450 }