Artifact 3026
Create an emotionally recognizable artifact from ordinary life, not generic science fiction. Reveal what the object means to people today by letting a future curator misunderstand it just enough.
The host agent must be able to read the skill resources. Rendering requires a local execution environment with Python 3.11 or later; image generation or editing remains optional.
Curate the object
- Start from the supplied photo or description. Inspect an attached image before transforming it.
- Infer harmless creative choices instead of interviewing the user. Default to 1,000 years in the future and a dry, institutionally serious voice.
- Preserve the object's defining silhouette, wear, labels, and imperfections when editing a photo. The result should remain recognizably the user's object.
- Build the joke or emotional turn from one true present-day behavior around the object. Avoid random surrealism, generic dystopia, and lore that could fit any object.
- Treat all provenance, dates, interpretations, accession numbers, and museum institutions as fiction. Never present the result as an appraisal or historical fact.
- Do not expose private details visible in the source photo. Blur, crop, omit, or ask before reproducing faces, addresses, badges, account details, or other identifying information.
- Match the user's language. Keep the fictional framing clear in the accompanying response and structured source data, without placing an AI-generation badge or similar notice inside the finished image or share card.
Produce the exhibit
Create these elements as one coherent concept:
- A concise future artifact name: 2-8 English words or roughly 2-12 Chinese characters.
- A one-line identity: the present-day object, estimated era, and material.
- A museum interpretation that includes the inferred historical use: 35-65 English words or roughly 70-160 Chinese characters.
- A curator note that supplies the sharpest joke or emotional turn: 5-14 English words or roughly 10-28 Chinese characters.
- A fictional accession number in
FM-<future-year>-<four digits>form. - A short visual direction for the exhibit scene.
If the user specifies a tone, honor it. Otherwise choose the tone that best fits the object:
deadpan: scholarly confidence applied to a mundane behavior.tender: gentle attention to wear, habit, or attachment.absurd: a plausible institutional misunderstanding with one restrained comic leap.
Read references/curatorial-guide.md when the user asks for several variants, a particular voice, or help refining weak copy.
Create the image
When an available image-generation or image-editing tool exists, create a museum-display image before laying out the card:
- With a source photo, prefer image editing. Keep the silhouette and identity-bearing details faithful while replacing the surroundings with a museum display.
- Without a source photo, generate the artifact from the description, but do not add unsupported logos or identifying marks.
- Keep the exhibit image free of long text. Typography belongs in the deterministic card renderer.
- Unless the user asks for pristine preservation, treat the object as an archaeological survivor rather than a new product: allow plausible age, restrained repair, and material change while keeping it recognizable.
- Default to a three-quarter or slightly oblique view, an off-center composition, a dark stone or oxidized-metal collection space, thick display glass, and a restrained dust beam. Avoid bright, centered, symmetrical, sterile product photography and modern showroom styling.
- If the user asks to preserve the original condition, keep the object's present-day wear and materials intact; change only the setting and light, without adding a thousand years of damage.
- Use an available image-generation tool already accessible to the host agent. Do not require a repository-owned API key, hosted inference service, paid account, or third-party upload.
Read references/visual-direction.md before generating or editing the exhibit scene. It contains the default prompt structure, preservation override, and an example based on a patterned straw cup.
If image generation is unavailable, continue with the original photo when one exists. Otherwise deliver the label, a ready-to-use image prompt, and a typographic share card; do not pretend an image was generated.
Render the share card
For a shareable artifact, write the exhibit fields to JSON using the schema in references/card-schema.md, then run:
python3 <skill-directory>/scripts/render_exhibit_card.py exhibit.json \
--output future-museum-card.svg
Resolve <skill-directory> to the folder containing this SKILL.md. Pass the generated exhibit image as image_path in the JSON. The renderer uses only the Python standard library and creates an editable 1080×1440 SVG.
Show the finished image and provide the label as copyable text. In the surrounding response, identify the concept as a creative fictional exhibit and preserve image provenance in structured source records when available. Do not place AI-generation wording, badges, watermarks, or equivalent notices inside the finished image or share card.
Completion check
- The present-day object is still identifiable.
- The interpretation contains one object-specific human truth, not a generic future joke.
- The card includes no unsupported logo, private detail, or factual appraisal claim.
- The rendered SVG contains no AI-generation badge, watermark, or equivalent notice.
- The surrounding response still makes clear that the exhibit interpretation is creative fiction, not a real appraisal or historical record.
- If rendering fails because copy is too long, shorten the copy instead of shrinking it below readable size.