Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add bozhouDev/video-skills-toolkit --skill minimax-voice-directorgit clone --depth 1 https://github.com/bozhouDev/video-skills-toolkitWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/bozhoudev/video-skills-toolkit/minimax-voice-director)<a href="https://agentmods.dev/skills/bozhoudev/video-skills-toolkit/minimax-voice-director"><img src="https://agentmods.dev/badge/skills/bozhoudev/video-skills-toolkit/minimax-voice-director/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/bozhoudev/video-skills-toolkit/minimax-voice-director"><img src="https://agentmods.dev/badge/skills/bozhoudev/video-skills-toolkit/minimax-voice-director.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00099 | $0.01415 |
| Opus 5 | $0.00049 | $0.00707 |
| Sonnet 5 | $0.00020 | $0.00283 |
| Haiku 4.5 | $0.00010 | $0.00142 |
Grade A, and why
minimax-voice-director scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 67 lines — stays where its author put it; the contents beside it link to each section on GitHub.
MiniMax 声音导演
把口播稿变成可审批、可复现、可局部返工的声音表演。不得把本 skill 缩减为“文本进、MP3 出”的单次 TTS 调用。
路由边界
- 用户明确选择 MiniMax 配音，或当前项目已经用 MiniMax：进入三阶段主流程。
- 用户只说 TTS、旁白、自己的声音，但没有选 MiniMax：不自动替换引擎，先根据上下文路由到合适的本地或云端工作流。
- 用户要 MiniMax Voice ID、声音克隆、声音设计或旧函数兼容：读
references/runtime.md，该分支不自动触发配音主流程。 - 音乐、BGM、歌曲或 cover：改用同级
music（ElevenLabs Music API）。 - 不因其他 TTS 失败而静默转用 MiniMax；先说明云端费用、声音差异和数据上传。
三阶段主流程
1. 先导演，再审批
- 定义原稿和字幕文本真值，先处理产品名、人名、数字、缩写和多音字。
- 读
references/directing.md设计全片基线、言语动作、潜台词、情绪弧线、节奏曲线、焦点词、语调和必要的声音事件。 - 读
references/schema.md，从assets/voice-direction.template.yaml创建work/tts/voice-direction.yaml。平台私有标签不得出现在导演中间层。 - 运行
scripts/validate_direction.py和scripts/compile_direction.py，产生审查视图、lint、MiniMax manifest 和干净字幕稿。 - 用独立审查回合检查导演逻辑，重点展示改写、强焦点、强转折、长停顿、声音事件、局部参数和多 Take 段落；通过后运行
scripts/mark_reviewed.py固化审查 hash。 - 停在这里等待用户批准。 用户没有明确批准时，不得运行
approve_direction.py，不得调用 MiniMax。 - 用户批准后运行
scripts/approve_direction.py，再重新编译。render_allowed必须为true。
2. 生成、选 Take，再批准音频
- 读
references/workflow.md和references/runtime.md，加载项目.env.r2，不输出任何密钥或 Voice ID。 - 运行
scripts/render_segments.py。它必须同时验证人工批准、lint 和内容 hash；任一失效就拒绝云端生成。 - 正式生成优先使用长连续块：开头约 40 秒一块，后续约 3 分钟一块；时长只是软目标，必须在句号或完整意群结尾切分，绝不从句子中间硬截。每块默认 1 个 Take，局部问题只重生局部。
- 听审候选并写入
take-selection.yaml，然后运行scripts/select_takes.py和scripts/finalize_voice.py。不得把原始候选或未重建停顿的拼接音频冒充最终成果。 - 在最终速度下验收开头 15 秒、一个中段、所有主要转折和收尾；检查发音、节奏、焦点、情绪连续性、音色漂移和声音标签执行。
- 再次停下等待用户批准最终音频。 获得批准后运行
scripts/approve_audio.py和scripts/publish_voice.py。
3. 只从定稿音频生成字幕
- 只对
audio_approved且 hash 匹配的最终 WAV 生成时间轴；M4A/MP3 是发布衍生物，不作为字幕时间基准。任何 Take、速度、裁剪或拼接变化都使旧时间轴失效。 - 使用
audio-to-subtitles生成 ASR 时间轴。该步骤需要 R2/MediaKit 凭证并会上传最终音频；使用项目已授权凭证，不打印密钥。 - ASR 只提供时间轴。用
work/tts/subtitle-source.txt回填最终显示文本，不显示呼吸、停顿和发音标记。 - 读
references/output-layout.md交付 SRT/VTT/JSON 和 raw ASR 证据，通过scripts/mark_subtitled.py把交付文件的 hash 绑定到已批准音频。数字人、口型和正式剪辑都必须以这份定稿音频为唯一时间基准。
引擎能力边界
导演稿使用 sound tags、局部参数或要求重音/语调控制时，读 references/capabilities.md。每个导演意图必须被标识为：
What ships with it
23 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- agents/openai.yaml 42 B
- assets/minimax_tts.py 24 KB runs code
- assets/voice-direction.template.yaml 3.2 KB
- references/capabilities.md 1.9 KB
- references/directing.md 3.6 KB
- references/output-layout.md 1.5 KB
- references/runtime.md 1.7 KB
- references/schema.md 2.9 KB
- references/workflow.md 5.2 KB
- scripts/approve_audio.py 2.8 KB runs code
- scripts/approve_direction.py 2.6 KB runs code
- scripts/compile_direction.py 10 KB runs code
- scripts/direction_contract.py 26 KB runs code
- scripts/finalize_segments.py 10 KB runs code
- scripts/finalize_voice.py 7.4 KB runs code
- scripts/mark_reviewed.py 2.1 KB runs code
- scripts/mark_subtitled.py 2.7 KB runs code
- scripts/publish_voice.py 8.5 KB runs code
- scripts/render_segments.py 8.8 KB runs code
- scripts/select_takes.py 3.5 KB runs code
- scripts/validate_direction.py 1.6 KB runs code
- scripts/validate_voiceover.py 2.9 KB runs code
- tests/test_voice_director.py 21 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 67 lines · 99 tokens per session scan A 43369d32b0d5
minimax-voice-director is a skill published in the GitHub repository bozhouDev/video-skills-toolkit (139 stars, last pushed 1mo ago), licensed MIT. It adds 99 tokens to every session and 1,415 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
video-strategy-orchestrator
Decides WHEN to use Remotion, HyperFrames, AI video generation, or a hybrid of all three. Use when building video production pipelines, choosing a video tool for a task, or when the user asks 'which video tool should I use?'. Evaluates input type, complexity, determinism needs, and output format to pick the right…
video-hyperframes
Hyperframes / Remotion-compatible continuous frame animation with autoplay support.
remocn
Build Remotion videos with remocn — copy-paste animation components and timeline-driven UI primitives from a shadcn registry. Use when composing a video or scene in a Remotion project, adding a single animation, transition, background, or UI-block sim, or reaching for a video-ready UI primitive (button, dialog…
video-hyperframes
A web-based sequence of video frames designed for Hyperframes or Remotion, with each frame presenting one visual idea.
minimax-cli
Nested swiss-knife reference for the MiniMax mmx CLI and the canonical MiniMax CLI procedure shipped with the TUI: install mmx-cli, discover the correct TUI-managed MiniMax preset/key slot without leaking secrets, match mainland vs international regions, and route image/video/music/TTS generation or one-shot shell…
videoagent-audio-studio
Tired of juggling multiple audio APIs? This skill gives you one-command access to TTS, music generation, sound effects, and voice cloning. Use when you want to generate any audio without managing multiple API keys.