Predictive Resource Allocation

io.github.farssaadeldin7-ux由 GitHub 账号 farssaadeldin7-ux 发布v0.1.3更新于 Nov 4, 2026

Predict where a render, simulation or training job will bottleneck, and what to change first.

概览

AI 生成的概览

面向 VFX 与 3D 工作室的插件,帮助助手在建议购买硬件前找出渲染、模拟或训练任务的瓶颈所在。

可以做什么
  • 预测渲染、模拟或训练任务会在哪里出现瓶颈,以及应先改动什么。
  • 按工作室流程先定位约束条件,再考虑硬件采购。
  • 部分工具可免费试用,其余工具需要付费许可。
  • 属于包含十四个插件的套件,也以 VS Code 扩展和 npm 包形式发布。
适用场景
适合 VFX 与 3D 工作室规划算力,或判断某个缓慢的渲染、模拟、训练任务应先换硬件还是先做其他调整。
运行要求
以本地进程方式运行(仅桌面端,stdio 传输),以 .mcpb 包分发。未声明认证、环境变量或请求头。除免费试用工具外,解锁其余工具需要付费许可。
安装前请注意
README 说明受许可保护的工具由计费服务控制,付费许可才能解锁其余功能,因此完整使用需要购买。README 也强调不得夸大能力:预测只是判断辅助,并非实测结果。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Predictive Resource Allocation,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

Plugin Suite

Fourteen domain-expert plugins for Claude Code and Cowork. Each one encodes how a specialist in that field actually works — the order they do things in, the question that splits the field fastest, and the point at which they say they do not know.

Browse the full catalogue with live pricing at codestudioplugin.com — plans for every plugin are listed on the pricing page. Each plugin's tool server includes free tools for evaluating it before buying; a paid licence unlocks the rest. Every plugin is also published to Open VSX as a VS Code extension — in Cursor, Windsurf, VSCodium or Gitpod, search code-studio in the Extensions panel and install.

Installable as a marketplace:

/plugin marketplace add farssaadeldin7-ux/claude-plugin-suite/plugin install diagnose-by-sound@plugin-suite

The plugins

PluginForThe judgement it encodes
diagnose-by-soundWorkshop service advisersCharacterise the noise before naming a cause; report calibrated confidence, not a guess
ghost-post-previewSocial managersName the most likely reason a post underperforms; never fabricate an engagement number
professor-mind-readerUniversity studentsAnswer at the level the rubric verb demands; find the sentence that satisfies each criterion
five-minute-fluencyCompetitive playersThree changes maximum, ranked by rating gain per unit of practice
trail-splitExpedition leadersEvery critical system has a named owner and a stated backup, or the plan is not finished
podcast-video-studioContent agenciesA clip that needs context is disqualified however good it was in the room
customer-sales-supportSaaS foundersAn AI support agent is only as good as the docs; "I don't know — here's a person" is a first-class answer
haptic-feedback-mapperCreative studiosA haptic that fires for something ignorable is worse than the screen it replaced
mental-health-chatbotTelehealth platformsSupport, not crisis intervention — escalation is always on and never a severity judgement
neural-link-intention-layerUI/UX designersEliminate, then batch, then automate — in that order
generative-digital-twinStudio headsA style is defined by what you never do; the "never" list matters most
emotional-resonance-analyzerAdvertising and marketing video editorsFlat stretches are the problem, not low stretches; real retention data always wins
code-to-visual-interpreterCreative technologistsStructure, modulation, surface — and a seeded PRNG or the piece is unreviewable
predictive-resource-allocationVFX and 3D studiosFind the binding constraint first; buying hardware is the last remedy, not the first

Audiences, the skill each user must bring, and the monetisation shape are in docs/AUDIENCES.md.

Design rules

These are enforced by review, and partly by scripts/validate.mjs.

  1. No overclaiming. If the premise implies a capability that does not exist — reading minds, measuring an audience that was never asked, predicting a number from text — the plugin says so in its first paragraph and reframes as what it honestly is. A user who trusts a meaningless number is worse off than one who got nothing.
  2. Thresholds, not vibes. "Under 15% separation, do not present a winner" beats "consider carefully". Every judgement in a skill should be checkable.
  3. The limits section is mandatory. Every skill states what it cannot answer and when to say so.
  4. References carry the tables. SKILL.md holds the procedure; lookup tables, taxonomies and worked examples live in references/ and are pointed at from the step where they become relevant.
  5. British spelling. No emoji, no hype vocabulary. The validator warns on both.

Layout

.claude-plugin/marketplace.json   the marketplace manifestplugins/<id>/                     one directory per plugin  .claude-plugin/plugin.json      plugin manifest  skills/<id>/SKILL.md            the procedure  skills/<id>/references/*.md     tables, taxonomies, worked examples  .mcp.json, mcp/                 only where a plugin ships a tool serverpackages/suite-runtime/           shared MCP + licensing runtime, vendored into pluginsservices/billing/                 the Stripe-backed licensing service the runtime talks toscripts/                          validate, vendor, builddocs/                             architecture, audiences, licensing

Working on it

bash
node scripts/validate.mjs        # structure, manifests, frontmatter, dead referencesnode scripts/vendor-runtime.mjs  # after editing packages/suite-runtimenode scripts/build.mjs           # .plugin archives into dist/node scripts/build-vsix.mjs      # VS Code extensions staged into dist/vsix/

CI runs all of these on every push. See docs/ARCHITECTURE.md for how to add a plugin, and docs/LICENSING.md for the free/paid split.

Publishing to Open VSX

Each plugin also ships as a VS Code extension: a thin extension.cjs that registers the plugin's MCP tool server through VS Code's MCP server definition provider API (VS Code 1.102+), with the whole zero-dependency mcp/ directory bundled inside the .vsix. The skills ship too: the bundled server publishes each one as an MCP prompt, which VS Code surfaces as a slash command (/mcp.<plugin>.<skill>). Listing is open; the skill body itself requires an active paid licence, same as the gated tools, since the .vsix is a public download.

One-time setup, in this order:

  1. Create an Eclipse Foundation account and log in to open-vsx.org with it.
  2. Sign the Publisher Agreement (open-vsx.org → your avatar → Settings → the agreement prompt).
  3. Create an access token (Settings → Access Tokens) and store it: gh secret set OVSX_PAT --repo farssaadeldin7-ux/claude-plugin-suite

Then publish from the Actions tab: run the publish-openvsx workflow (optionally naming specific plugin ids). It packages every extension with vsce, creates the code-studio namespace if needed, and uploads each .vsix with ovsx, skipping versions that are already published. To build locally instead: node scripts/build-vsix.mjs --package.

The plugins also publish to npm as @farseman/<id> packages with the publish-npm workflow (runnable as npx @farseman/<id>; needs the NPM_TOKEN repository secret described at the top of .github/workflows/publish-npm.yml), and to the official MCP Registry with the publish-mcp-registry workflow: scripts/build-mcpb.mjs packages each one as an .mcpb bundle attached to a GitHub release, and the workflow submits each listing under the io.github.farssaadeldin7-ux namespace, authenticated with the job's own GitHub OIDC token — no stored secrets.

The same extensions publish to Microsoft's VS Code Marketplace with the publish-vsmarketplace workflow. That marketplace has no trusted publishing, so it needs a Personal Access Token stored as the VSCE_PAT repository secret — the one-time setup is described at the top of .github/workflows/publish-vsmarketplace.yml.

Status

Version 0.1.3. All fourteen plugins ship working MCP servers built on the shared runtime, each gating its licensed tools behind the billing service and keeping open tools for evaluation before buying.

Licence

Source-available, not open source. The code is published so it can be read and evaluated; publication grants no rights to copy, modify, redistribute or create derivative works from it. Paid plugin tiers are governed by the commercial terms shown at the point of purchase. See LICENSE.

来源:README.md,提交 815489a

工具

0
SourceWeft 曾在隔离沙箱中以无凭据方式启动该软件包一次,以列出其工具,但未能完成:该软件包类型无法在沙箱中运行。

版本历史

1
  1. v0.1.3最新Oct 21, 2026

评分

已满足适用检查项的 40%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

    请在仓库或注册中心条目中声明许可证。