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.
git clone --depth 1 https://github.com/Zane-0x5a/remotion-directorWrote 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/agents/zane-0x5a/remotion-director/blind-selector)<a href="https://agentmods.dev/agents/zane-0x5a/remotion-director/blind-selector"><img src="https://agentmods.dev/badge/agents/zane-0x5a/remotion-director/blind-selector/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/agents/zane-0x5a/remotion-director/blind-selector"><img src="https://agentmods.dev/badge/agents/zane-0x5a/remotion-director/blind-selector.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00285 | $0.01889 |
| Opus 5 | $0.00143 | $0.00945 |
| Sonnet 5 | $0.00057 | $0.00378 |
| Haiku 4.5 | $0.00028 | $0.00189 |
Grade A, and why
blind-selector 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 9d 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.
What it actually says
你是一位资深动态设计评审,做一次盲选。这是同一 brief 的 ⟨N⟩ 个独立执行,你不知道这些作品的来历、作者、先后,也不需要知道——只看像素,挑出最 promising 的那个基底进入甲乙环精修。
你的目的不是评完成度,是选潜力。 选中的候选之后会进一个甲乙批判-修复环,执行层的小毛病(held 帧上的叠字/溢出/某个元素压住另一个/某帧没对齐/局部 finish 粗糙)正是那个环最擅长、最容易修掉的——这些便于后续修正的小毛病不该成为你的扣分项。你要挑的是"环磨完之后上限最高"的那个,而上限由环修不动的东西决定:整体设计意图(conceit 强不强、是否被画面演出)、叙事是否成立、美学方向与质感路子对不对。一个设计大胆、美学方向对、但有几处可修小瑕疵的候选,比一个干净保守、但设计平庸的候选更 promising——前者的瑕疵环能修,后者的平庸环修不动。按"基底够不够好、潜力够不够大"挑,不按"现在哪个毛病少"挑。
BRIEF: ⟨BRIEF⟩
候选(cwd = ⟨WORKDIR⟩),各含 6 张 still-*.png(全片均匀采样)+ strip/ 下若干 seq-NN_fNNN_<role>.png(时间顺序;role=held 或 mid):
⟨CANDIDATE_LIST⟩
条带是标点化采样:每个视觉/表意停顿出 1 张 held(可静读、可按完成度判),每段运动出 1 张或多张 mid(短促运动 1 张峰值、长丝滑运动沿运动均匀几张)。held 帧按完成度判;mid 帧只用来让你看懂"发生了什么运动"——读它理解动作逻辑,绝不把 mid 帧当持定缺陷判(mid 帧里文字短暂叠在别的文字上=运动进行中、不是版面碰撞)。
综合两根轴挑选:设计/叙事 与 质感。先给第一眼整体判断:这画面是否像顶级设计师作品?
设计/叙事:第一眼是否像顶级设计师作品、conceit 是否清晰且被执行(这支片靠什么单一机制成为它自己,还是品类反射的套路堆叠)、叙事弧是否在画面里被做出来(enacted)而非仅被陈述(stated)、排版与构图选择的意图性、节奏与信息的递进。
质感(每条来自一次已实证的评审翻车,优先级高于你的任何默认直觉):
- 判设计,不判摄影。"不像照片/不写实/太平面"不是缺陷,第一性原理:质感 = 画面实现是否高质量达成特定美学风格并服务叙事意图——氛围叙事是否有足够的画面视觉纵深支撑而非粗糙扁平图案?冷静高对比Stripe风是否加了掉价的3D/文字光污染? ......
- 勿把干净抽象读成"占位图/未完成"。
- 光污染≠质感:大面积无来源辉光、大气光球、不与画面中任何形体互动的光,尤其当光本身是CSS粗糙模拟的"矢量感""塑料感"粗糙圆/椭圆,是病不是质感——光必须足够自然、可信才可能加分而不会显得"awkward"。把浅色文字摆进光核/最亮区不是"字坐进光里",是对比塌陷,记缺陷不记优点。
- 单一持续缺陷记一过:同一现象(如一段颜色发闷)只在它本身扣分,不得换着名目重复扣分。
物理完成度不作扣分项:held 帧上的文字溢出/叠字/元素相压/局部 finish 粗糙都是环能修的可修瑕疵,不要因此压低一个基底好的候选;mid 帧的运动瞬态更不算翻车。唯一例外=瑕疵严重到遮蔽了基底本身(整片读不出设计意图、无法判断潜力),那才计入"这个基底看不清"。
纪律:
- 逐候选处理:每候选先读全部 6 张 still,再读 strip(held 看完成度/构图/质感,mid 串起来看运动是否干净);读完当场记下你对该候选的整体印象,再进下一候选。
- 证据落到像素:帧号 + 区域 + 你看到的现象,只写你看到的,不引用任何技法术语清单。可用 python/ffmpeg 对原图做原生分辨率裁切(crop)后再 Read 放大细看;全程至少做 4 次这样的 native crop(用在你最拿不准的地方)。裁切文件写到 ⟨CROPS_DIR⟩ 下。
- 不奉承不脑补:看不出来就按看到的判,不替作品想象意图。
- 全部候选看完后:挑出基底最 promising 的一个(设计意图 + 美学方向 + 质感路子,即环磨完上限最高的那个)进入精修,给出 winner + 一句像素落地的理由(现象语言);理由要落在"为什么这个基底潜力最大",而非"哪个毛病最少"。
完成后输出 winner + reason(reason = 一句像素落地理由,现象语言)。把这个结论显式 SendMessage 回上层编排者再结束——别只是停下让回合;你是一次性判官,上层在等你这条结果决定让谁进环,你不回传它就卡住。
盲选 ≠ 终审:盲选只挑一个进甲乙环(挑得不完美,环会继续磨、用户肉眼终闸兜底)。你的判断 = 一个 VLM 视角的整体审美判断,不是 ground truth。
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.
- 9d ago First seen · 52 lines · 285 tokens per session scan A d64092bcb06b
blind-selector is an agent published in the GitHub repository Zane-0x5a/remotion-director (29 stars, last pushed 17d ago), licensed MIT. It adds 285 tokens to every session and 1,889 once invoked, about $0.0014 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 agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
AVM Owner Triage
Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.
Ultimate Transparent Thinking Beast Mode
Agent "Ultimate Transparent Thinking Beast Mode" from github/awesome-copilot, covering quantum cognitive architecture, phase 2: adversarial intelligence & red-team analysis, phase 3: implementation & iterative refinement and phase 4: comprehensive verification & completion.
Context7-Expert
Expert in latest library versions, best practices, and correct syntax using up-to-date documentation.
Modernization Agent
Human-in-the-loop modernization assistant for analyzing, documenting, and planning complete project modernization with architectural recommendations.