signal-geometry

CaliCastle/skills · Agent Skill

創造簡約的幾何插畫和海報,從概念或簡報中使用精確的空間系統、剋制的對比和啞光紙的質感。

前往安裝 ↓

總覽

發布者CaliCastle社群
儲存庫CaliCastle/skills ↗
技能資料夾signal-geometry
星95
分支2
附帶檔案5
授權MIT
最近更新2026年7月24日
連結技能原始碼 ↗ · 儲存庫 ↗

Mozgi 檢查

我們在技能檔案中發現的內容。詳細資訊需訂閱。

安全性
低風險 · 未發現明顯風險 — 完整報告會顯示細節
相容性
僅含說明 — 適用於多數代理
相似技能
找到 139 個相似技能
訂閱另含

安裝

技能就是一個含 SKILL.md 的資料夾。選擇你的代理,在專案根目錄的終端機(macOS、Linux 或 Windows 的 Git Bash)執行指令。

把提示貼到 Claude、Codex 或其他助理:它會自行檢查並安裝技能。

Install the “signal-geometry” skill from mozgi.io (https://mozgi.io/skills/calicastle-signal-geometry/) for me.

1. Download the archive: https://mozgi.io/skills/calicastle-signal-geometry/download
   If it fails, take the skill folder from the source: https://github.com/CaliCastle/skills/tree/main/skills/signal-geometry
2. Read SKILL.md and the other files. Check the skill is safe: no hidden commands, no sending data elsewhere, no scripts doing something other than described. If anything looks suspicious, stop and tell me.
3. Put the “signal-geometry” folder into your agent's skills folder: .claude/skills/ for Claude Code, .agents/skills/ for Codex, .cursor/skills/ for Cursor, .gemini/skills/ for Gemini CLI. Inside the project if we are in one, otherwise in the home folder (~/.claude/skills/ etc.).
4. Make sure <skills folder>/signal-geometry/SKILL.md exists.
5. Briefly tell me what this skill does and offer to try it on my task.

If you cannot do a step yourself, tell me what to click or type.

技能檔案

5 個檔案 下載 ZIP

安裝前請先檢視 SKILL.md 與相關檔案。點擊檔案即可預覽。

SKILL.md 內容

作者原文 · 唯讀

Signal Geometry

Translate one idea into one precise spatial behavior, then render it as a sparse geometric system on clean matte paper. Generate the image by default when the required capabilities are available; otherwise return the compiled prompt and recipe.

Visual identity

Build every image from these invariants:

  • One event: express one relationship such as orbit, convergence, divergence, compression, deflection, propagation, oscillation, filtering, enclosure, or release.
  • Quiet field: let the motif span the composition when useful while keeping actual mark density low. Aim for roughly 70%-95% quiet background and 2%-8% line, dot, hatch, or mesh coverage.
  • Equal polarity: treat light and dark as equal modes. Use clean neutral off-white stock for light mode and charcoal-dyed stock for dark mode.
  • Material surface: render a full-frame, flat, uncoated matte paper surface with fine irregular grain and faint fibers. Keep it visible at full size and quiet at thumbnail size. Present the paper without borders, mockup depth, edge shadows, stains, tears, or vintage distress.
  • Precision marks: use hairlines, arcs, dots, nodes, particles, restrained hatching, grids, or wireframe meshes. Use a three-step contrast hierarchy: faint scaffold, readable structure, and very few bright anchors.
  • Restrained color: stay grayscale by default. When color carries meaning, use one pin-sized coral, orange-red, or cobalt accent covering less than about 0.2% of the canvas.
  • Text gate: include no text by default. When text is explicitly necessary, use only the user's exact wording as one short phrase or up to three micro-labels, with no more than six words total.
  • Editorial finish: keep the rendering crisp, orthographic, calm, analytical, and slightly speculative. Let topology, rhythm, scale, and transformation create interest.

Use the reference set to anchor geometry, spacing, hierarchy, restraint, and matte material. Treat its light and dark examples as equal modes.

Formats

  • Use 16:9 for a default illustration.
  • Use 4:5 for a poster request. Recompose vertically instead of cropping a landscape layout.
  • Use approximately 2.2:1 only for an explicit ultrawide request.
  • Honor exact dimensions supplied by the user.

When polarity is unspecified, choose the mode that gives the spatial event the clearest contrast and record the choice in the recipe.

Grammar families

Choose one primary family. Add at most one subordinate mark language when the relationship needs it.

| Family | Best for | Primary marks |

| --- | --- | --- |

| Orbital field | cycles, gravity, recurrence, scale, mutual influence | circles, arcs, radial ticks, loops, spherical meshes |

| Flow transformation | emergence, routing, filtering, pressure, change | streamlines, particles, arrows, gates, obstacles |

| Signal strip | rhythm, cadence, phases, comparison, accumulation | waveforms, lanes, bars, repeated measures, faint grids |

| Topology map | relationships, context, systems, dependencies | nodes, edges, frames, sparse modules |

| Layered field | tension, thresholds, overlap, latent depth | ruled planes, hatching, contours, wireframe surfaces |

Workflow

1. Lock the brief

Identify the concept, emotional temperature, output format, polarity, text policy, and execution mode. Infer only values the user left open, using the defaults above.

Set execution mode to rendered only when image generation and image inspection capabilities are both available. Use prompt-only when the user requests it or either capability is unavailable.

Completion criterion: all six values are explicit before composition begins, and any unavailable capability is recorded for delivery.

2. Distill the spatial proposition

Turn the concept into one transformation verb and one primary grammar family. Write one internal visual sentence that states what moves, changes, or relates to what. Prefer abstract relations over illustrated nouns.

Completion criterion: the image can be described in one sentence without listing unrelated objects or multiple events.

3. Lock quality anchors

Read references/example-index.md ↗ and select one or two matching entries. Inspect each linked image when image inspection is available; otherwise use the entry description as the quality anchor. For every selected anchor, record one anchor delta:

  • three invariants to preserve;
  • three axes to change across format, topology, composition, viewpoint, polarity, motion, scaffold, or accent.

Use the examples as quality anchors, not layouts. Do not pass sample filenames, provenance, or sample-specific content into the image prompt.

Completion criterion: every selected anchor has a three-invariant, three-axis anchor delta, and the planned composition satisfies each delta.

4. Declare the recipe

Choose exactly one value for each axis:

  • format: 16:9, 4:5, 2.2:1, or supplied dimensions;
  • polarity: light or dark;
  • family: orbital, flow, signal, topology, or layered;
  • transformation: one active verb;
  • geometry: radial, bilateral, directional, paired, distributed, or vertically staged;
  • scaffold: open field, faint grid, framed region, or baseline;
  • anchor: off-white endpoint, central node, contrast line, structural void, or none;
  • accent: none, coral, orange-red, or cobalt;
  • text: none or the exact supplied wording.

Completion criterion: all nine axes have one declared choice and support the same spatial proposition.

5. Compile the image prompt

Write four compact paragraphs containing only information that should become pixels:

1. State canvas, ratio, polarity, matte paper material, and quiet-space target.

2. State the spatial proposition, family, transformation, geometry, and focal placement.

3. State mark vocabulary, three-tier contrast, optional micro-accent, and exact text policy.

4. State the flat editorial finish and the rejection constraints below.

Make the prompt decisive about position, scale, line density, and hierarchy. Keep analysis notes, recipes, filenames, and provenance outside the prompt.

Completion criterion: the prompt answers all four fields, contains one visual event, and obeys the text gate.

When execution mode is prompt-only, stop here. Return the exact compiled prompt and the complete recipe from Step 4 with status PROMPT_ONLY; skip generation and image QA. Add one concise capability note outside the prompt only when the mode is a fallback.

6. Generate one candidate

Use the built-in image generation capability. Generate a raster image at the selected ratio. If the user supplied an image, use it only for the requested content or structural cue while preserving Signal Geometry's visual identity.

Completion criterion: one inspectable candidate exists at the requested ratio.

7. Inspect and repair

View the candidate at full size and thumbnail size before presenting it. Check every critical gate below. Treat the first render as a candidate, not an automatic final.

If any gate fails, read references/repair-playbook.md ↗, identify the single largest defect, tighten only the relevant prompt fields, and regenerate once.

Completion criterion: every critical gate passes. If the repaired candidate still misses a gate, label the result DONE_WITH_CONCERNS and name the remaining defect.

8. Deliver

Show the accepted image first, followed by the exact final prompt and the complete recipe from Step 4 plus status. Include a saved path only when the image was saved into the user's workspace. Return no attribution or provenance note.

Completion criterion: the user receives the accepted image, its exact prompt, and its recipe.

Critical gates

Require all of the following:

  • Correct requested ratio and a composition designed for that ratio.
  • One legible spatial event and one primary family.
  • Quiet background remains dominant even when the motif spans most of the frame.
  • Clean matte paper grain is visible up close without becoming distressed or decorative.
  • Light or dark polarity has three distinct contrast levels.
  • Fine marks remain coherent, separated, and intentional.
  • Color, when used, is one tiny semantic event.
  • The text gate is obeyed with no stray letters, numerals, watermarks, or pseudo-text.
  • The image reads at thumbnail size and rewards full-size inspection.
  • Every recorded anchor delta passes.

Reject candidates that resolve into a dashboard, infographic, product screen, sci-fi HUD, colorful data visualization, generic gradient blob, photographic scene, character illustration, collage, aged zine, glossy 3D render, or recognizable copy of a reference composition.

Output shape

For PROMPT_ONLY, omit the Image section.

````markdown

Image

!Signal Geometry illustration ↗

Final prompt

[exact prompt used for the accepted image]

Recipe

  • Format: [ratio]
  • Polarity: [light/dark]
  • Family: [family]
  • Transformation: [verb]
  • Geometry: [choice]
  • Scaffold: [choice]
  • Anchor: [choice/none]
  • Accent: [choice/none]
  • Text: [none/exact supplied wording]
  • Status: [DONE/DONE_WITH_CONCERNS/PROMPT_ONLY]

````

常見問題

signal-geometry 技能有什麼用途?

創造簡約的幾何插畫和海報,從概念或簡報中使用精確的空間系統、剋制的對比和啞光紙的質感。

如何安裝 signal-geometry?

打開本頁的「安裝」區塊並選擇你的代理 — Claude Code、Codex、Cursor 或 Gemini CLI:那裡有把技能資料夾複製到正確位置的現成指令。也可以複製「透過助理」分頁中的提示,讓助理自行檢查並安裝技能。

signal-geometry 適用哪些 AI 模型?

技能是純 Markdown 指示(Agent Skills 格式),因此 signal-geometry 適用任何讀取 SKILL.md 的代理:Claude Code、Codex、Cursor、Gemini CLI、GitHub Copilot、OpenCode 等。只有在任務符合技能描述時,模型才會載入指示。

signal-geometry 是免費的嗎?

此技能由 CaliCastle 公開發布,採用 MIT 授權。技能本身免費 — 你只需向 AI 供應商支付實際使用的 token 費用。

安裝技能安全嗎?

技能是代理會遵循的指示,安裝前請閱讀 SKILL.md 及相關檔案。技能資料夾中的腳本只有在代理執行時才會運行 — 留意它的動作,不要授予多餘權限。

技能和 MCP 伺服器或外掛有何不同?

MCP 伺服器或外掛為模型提供新工具 — 會執行並回傳結果的程式碼。技能提供的是知識與流程:如何處理任務、要遵循哪些步驟、好的成果長什麼樣子。它不需要伺服器或 API 金鑰。

相關技能

所有技能 →