jnMetaCode/agency-orchestrator is a workflow system that lets people assemble teams of specialized AI roles from a natural-language request and run them against defined tasks and acceptance criteria. It is intended for individuals who want AI collaborators to produce research, content, products or other deliverables through web, desktop, Docker or command-line interfaces. The catalogue add-ons configure the agent roles and workflows used by the orchestrator.
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 jnMetaCode/agency-orchestrator --skill shortfilm-promptgit clone --depth 1 https://github.com/jnMetaCode/agency-orchestratorWrote 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/jnmetacode/agency-orchestrator/shortfilm-prompt)<a href="https://agentmods.dev/skills/jnmetacode/agency-orchestrator/shortfilm-prompt"><img src="https://agentmods.dev/badge/skills/jnmetacode/agency-orchestrator/shortfilm-prompt/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/jnmetacode/agency-orchestrator/shortfilm-prompt"><img src="https://agentmods.dev/badge/skills/jnmetacode/agency-orchestrator/shortfilm-prompt.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 | $0.00066 | $0.01611 |
| Opus 5 | $0.00033 | $0.00805 |
| Sonnet 5 | $0.00013 | $0.00322 |
| Haiku 4.5 | $0.00007 | $0.00161 |
Grade A, and why
shortfilm-prompt 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 — 85 lines — stays where its author put it; the contents beside it link to each section on GitHub.
5 段式视频提示词方法论
改编自姊妹项目 ai-shortfilm-prompts（MIT，同作者）的
methodology.zh.md 与 cases.zh.md。上游那份是交互式 Claude Code Skill（会向用户提问、按需读模板文件）；
这里是单轮非交互改编：你没有工具、也不要提问，直接按下面的规则一次写完。
核心心法一句话：像写分镜脚本一样写提示词，不要像写文学描述。 把场景按时间切片， 每一片交代「动作 / 镜头 / 特效 / 声音」四件事。
结构：5 段，顺序不许乱
1. 核心主题 ← 3-6 个 tag，用 | 分隔，从"画面类型 → 题材 → 美学风格"层层递进
2. 人物与设定 ← 面部 / 服装 / 场景，各一行
3. 氛围与画质 ← 视觉基调 / 色彩与影调 / 声音
4. 运镜规则 ← 镜头方式 / 角度 / 呼吸感
5. 分镜 ← 按秒切片，每片写清动作与声音
段 1 · 核心主题
一行标签。真实感锚点（电影级质感、超写实、杜绝游戏 CG 感）＋ 美学定调（暗黑写实、原子朋克、 废土工业、复古港风）＋ 题材 ＋ 情绪。"电影感""高级""震撼"单独出现即为不合格——它们不是描述，是形容词垃圾。
段 2 · 人物与设定
面部 / 服装 / 场景三行。服装写质地不写品类（"哑光黑色皮质""磨砂金属"，不是"夹克""腰带"）。 至少两处瑕疵：磨损、尘土、汗渍、伤痕——干净等于假。
有参考图时，把脸锁死，否则镜与镜之间人会漂：
面部：参照上传图片，五官、脸型、发型 100% 还原，杜绝美化。保留伤口 / 疤痕 / 瑕疵。
段 3 · 氛围与画质
必须写具体摄影机型号 + 镜头型号 + 焦段/光圈 + 色调 + 光源。给 AI 一个具体的视觉锚点， 比十个形容词管用：史诗镜头用 IMAX 电影摄影机 + Panavision C 系列（35mm，f/4）， 硬核科幻用 Sony Venice + Canon K-35。
色调必须在这里定死并逐镜一致。AI 生成视频色彩位深低，后期调色一推就断层出噪点—— 色要在提示词里修，不要指望在剪辑软件里修。
声音固定写明，否则模型会自作主张配乐：
声音：无配乐，仅保留现场音。
段 4 · 运镜规则
写精确：相机 + 镜头、景别（远/中/近）、角度（低/高/平/过肩）、运动（推/环绕/跟/FPV）、 节奏（极慢/平稳/突发/反应式）。"围绕主角来一段有张力的运镜"是废话， "中近景，低角度偏左 30°，顺时针环绕，极慢，带 0.1 秒反应式微抖"才是指令。
呼吸感那一句永远带（只有刻意要僵硬的定镜/俯拍才去掉）：
手持拍摄，全程保持极其轻微的、如呼吸般的镜头浮动，增强临场感。
段 5 · 分镜
按秒切片（0-2s / 2-4s / 4-5s……），每片写清这一秒发生什么 + 声音。
硬规则（这几条踩了就是废片，钱直接烧掉）
- 不出现任何影视 IP 名、角色名、明星姓名。 厂商的版权过滤会在提交这一步就拦。 改写成设计语言：「钢铁侠红金战甲」→「原子朋克复古未来感红金战衣」。
- 单镜不超过 15 秒。 甜区按场景类型 5–15s，硬撑 30s 一镜到底会在中段画质崩坏。 更长的序列拆成多镜，靠续写或剪辑接。
- 不要写空泛词。 每段都要有具体名词。
- 控制总字数。 写长了模型注意力散，而且会撞厂商字数上限直接提交失败。 要压就删冗余形容词、合并相关短句、缩写场景行——不要删具体型号和分镜。
- 结尾要克制。 不要在最后一镜堆爆炸、光爆、粒子。安静收尾更高级：
人停在那里，站着，呼吸，在烟尘里，配一个远处的元素（一颗划过天际的流星）。
没有台词。没有爆炸。没有光芒万丈。只有身穿不完整战衣的人站在原地。
- 多镜之间靠运动方向连贯性衔接，不是靠对齐静帧：这一镜主体从画面右侧被推出， 下一镜就从右侧进画。设计镜头时刻意安排出画 / 入画的边。
- 动态镜头（变身、打斗）不垫图。 参考图会把它自己的画风（二次元、手绘、CG 渲染） 带进输出。参考图只用于角色/场景连贯性，且必须是高清无滤镜的写实素材。
心态
一条提示词是一张抽卡券，不是确定性配方。同一条提示词重抽出的结果天然不同—— 高难度镜头抽 20+ 次、简单镜头 2–3 次是常态。把功夫下在挑输出上，别指望把提示词打磨到完美。
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 · 85 lines · 66 tokens per session scan A 3197f750d1df
shortfilm-prompt is a skill published in the GitHub repository jnMetaCode/agency-orchestrator (2,232 stars, last pushed 2d ago), licensed Apache-2.0. It adds 66 tokens to every session and 1,611 once invoked, about $0.0003 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
ai4j-app-builder
Use this skill when helping users build applications with AI4J in their own Java or Spring Boot projects, including first chat, streaming, tool/function calls, MCP, RAG, memory, Agent runtime, Coding Agent CLI embedding, FlowGram integration, provider configuration, dependency selection, and troubleshooting. It guides…
ask-user-collaboration
Ask the application user concise structured questions when missing information blocks a safe AI4J agent action.
use-v4-flash-worker
Use only for an existing Codex 0.148.x-or-older DeepSeek-backed v4flashworker installation; contains the legacy plaintext-Hook handoff, recovery, and data-boundary contract. On Codex 0.149.0 or later, direct cross-provider work to MixAgents Broker instead.
broker
Delegate bounded work from a Codex controller to configured workers on other model providers without changing the controller's provider. Use when cross-provider delegation, a named Broker route, or a lower-cost/specialized external worker is requested. Do not use for ordinary same-agent work.
code-review
A code-review checklist for examining code without rewriting it. It checks correctness, security, readability and performance, then labels findings by seriousness.
email-assistant
A read-only assistant for reviewing incoming Yandex or Gmail messages from a chat.