image-skillbook · diff

v0.1.0 to v0.1.0

35 added, 46 removed. Audit A to A.

---

name: image-skillbook

description: >-

- Discover, select, run, and evaluate a curated library of visual treatments for AI

- image generation and editing. Use when the user wants style recommendations, names

- a Skillbook recipe, wants several photos processed separately, or wants a sourced

- social prompt turned into a reusable workflow. Do not use for basic file conversion.

+ Compare Image Skillbook treatments and route an image request to the most suitable

+ installed style Skill. Use when the user wants help choosing an AI photo treatment,

+ wants several styles compared, or names Image Skillbook without choosing a style.

+ Do not use for ordinary image conversion or when a specific style Skill is already named.

+ license: MIT

metadata:

version: "0.1.0"

author: AlbertAZ1992

---

# Image Skillbook

- Route an image request to one focused visual Skill. This is a catalog and workflow,

- not a universal mega-prompt.

+ Help the user choose one focused visual treatment, then route execution to that style's own

+ instructions. This Skill is a selector and does not replace the independently installable

+ style Skills.

- ## Choose a style

+ ## Compare the treatments

| Style Skill | Choose when |

| --- | --- |

- | `editorial-painted-memory` | sparse acrylic shapes on warm paper |

- | `enamel-travel-keepsake` | a place should become a physical souvenir object |

- | `crayon-memory-postcard` | visible wax grain and naive travel-journal warmth |

- | `rubber-stamp-travel-journal` | location contours should become carved ink layers |

- | `painted-editorial-reconstruction` | the lower scene needs a broad painterly retelling |

- | `photo-doodle-story` | real subjects should interact with tiny line figures |

- | `impasto-miniature-world` | the scene should become sculptural thick-paint relief |

- For discovery, recommend at most three styles and give one source-specific reason for

- each. Do not generate until the user chooses unless they asked you to decide.

+ | `editorial-painted-memory` | the image needs sparse acrylic shapes on warm paper |

+ | `enamel-travel-keepsake` | a person or place should become a physical souvenir object |

+ | `crayon-memory-postcard` | wax grain and personal travel-journal warmth fit the subject |

+ | `rubber-stamp-travel-journal` | place-specific contours should become carved ink layers |

+ | `painted-editorial-reconstruction` | a scene needs a broad painterly retelling |

+ | `photo-doodle-story` | a person or pet can interact with tiny line figures |

+ | `impasto-miniature-world` | the scene can become sculptural thick-paint relief |

- For a named style, follow that installed Skill's instructions. Do not blend styles

- unless the user explicitly requests a hybrid. If the style Skill is unavailable,

- apply the matching contract in the table faithfully and state the limitation.

+ Inspect the supplied image and the user's intended mood, subject fidelity, layout, and output

+ use. Recommend at most three treatments. Give one image-specific reason and one material

+ trade-off for each recommendation.

- ## Shared workflow

+ Do not generate during comparison unless the user asked the router to choose and proceed.

- 1. Label each supplied image as edit target, reference, or compositing insert.

- 2. Keep one input per output unless the user explicitly asks for a combined result.

- 3. Preserve originals and write outputs non-destructively.

- 4. Use Direct mode for a proven creative treatment, Adaptive mode when the subject

- controls the design, and Production mode for exact geometry or source fidelity.

- 5. Compile only the selected style, image-specific facts, exact delivery constraints,

- and a short avoid list. Repetition can weaken image-model results.

- 6. Use the best available image generation or editing tool. If none is available,

- return a ready-to-paste prompt instead.

- 7. Inspect every result for subject identity, style specificity, text, extra objects,

- watermarks, dimensions, and separation between inputs.

+ ## Route the request

- When an unchanged photo region or exact split is required, generate expressive artwork

- separately and assemble the final canvas deterministically. Do not repeatedly ask an

- image model to fix measurable geometry.

+ When the user chooses a treatment:

- ## Contribute a style

+ 1. Check whether that style Skill's instructions are available in the current session.

+ 2. If available, follow that Skill exactly. Keep one input per output unless the user asks for

+ a combined composition.

+ 3. If unavailable, do not approximate the treatment from this comparison table. Return its

+ exact install command:

- Treat social-media prompt text as private intake unless its license permits republication.

- Record the author, URL, model, example, and reuse terms when available. Extract the

- visual technique, write an independent execution contract, and keep the raw prompt out

- of the public Skill when rights are unclear.

+ ```bash

+ npx skills add AlbertAZ1992/image-skillbook \

+ --skill <style-skill-name> --global --agent codex --yes

+ ```

- In the source repository, scan every Markdown file under `prompt-inbox/`, excluding

- `README.md` and `STATUS.md`. The user may create one file per prompt. Inventory before

- deduplicating, preserve every raw entry, and never delete one until its specific Skill

- has been reviewed and accepted.

+ 4. Ask the user to start a new agent session after installation, attach the image again, and

+ invoke `$<style-skill-name>`.

- New styles remain Draft until tested. Promote a style only after it works across at

- least three meaningfully different inputs and wins a human comparison.

+ Do not blend treatments unless the user explicitly requests a hybrid. In that case, explain

+ which visual decisions come from each Skill before generation.