Prompt Library: GitHub Markdown/text import and curated repo search

[lidge-jun]

Summary

Extend Prompt Library import so users can import prompt collections from GitHub Markdown files and curated public repositories, not only local files.

Requested behavior

1. GitHub source import

• Accept a GitHub URL pointing to a Markdown file.
• Accept a repo plus Markdown path form, so users can enter a GitHub address and an .md path without manually converting to raw URLs.
• Resolve GitHub blob URLs to raw Markdown internally.

2. Import UX

• 불러오기 should open a Prompt Library import popover/modal instead of opening Finder immediately.
• The popover should say to drag files here and include a GitHub URL/path input.
• While the popover is open, whole-screen drag/drop should be accepted and routed to the same import flow.
• Finder/file picker remains available as a secondary button inside the popover.

3. Curated repository search

• Add a small curated/public-repo search section for known prompt/image-generation repositories.
• Seed with PicoTrex/Awesome-Nano-Banana-images as the high-star Nano Banana image prompt/example source.
• Seed with aimikoda/nano-banana-pro-prompts as a Nano Banana Pro / Nano Banana 2 prompt source.
• Users should be able to pick a repo, browse/import Markdown prompt files, and save extracted prompts into the existing Prompt Library.

Suggested input formats

https://github.com/owner/repo/blob/main/path/to/prompts.md
https://github.com/owner/repo/tree/main/path/to/folder
owner/repo:path/to/prompts.md
owner/repo@branch:path/to/prompts.md

Acceptance criteria

• Clicking 불러오기 opens an import dialog/dropzone instead of Finder.
• Dragging a local .md / .txt anywhere over the app while the dialog is open imports through the dialog flow.
• Pasting a GitHub Markdown file URL can preview/import prompts.
• Curated repo quick picks include Nano Banana sources.
• Imported prompts include source tags such as github, repo name, and filename.
• Existing local file import behavior still works from inside the dialog.
• Tests cover dialog open behavior, GitHub URL parsing, host/path validation, and parser output.

Devlog

Plan: devlog/_plan/0.23.1-prompt-library-github-import/PRD.md

[lidge-jun]

Scope update: GitHub import must support plain text files as first-class sources too, not only Markdown. Accepted remote/file formats are now .md, .markdown, and .txt; .txt prompt extraction should split by clear separators such as blank-line blocks, --- separators, numbered headings, or one-prompt-per-line when lines are long enough. Devlog PRD updated accordingly.

[lidge-jun]

Oracle plan audit result: NEEDS_CHANGE, but direction is valid. The key recommendation is to reduce PR1 to import UX correction plus single GitHub file preview/import for .md/.markdown/.txt only. Folder browse, recursive indexing, global GitHub discovery, JSON datasets, and large curated-source search should be later PRs. PRD updated with: PR1-PR4 scope split, curated source registry fields, seed/candidate repos, source-file and prompt-candidate index schema, search ranking criteria, caps, and expanded security/parser tests. Devlog audit file: devlog/_plan/0.23.1-prompt-library-github-import/ORACLE-AUDIT.md

[lidge-jun]

GPT Image 2 research addendum

Added local planning note: devlog/_plan/0.23.1-prompt-library-github-import/GPT-IMAGE-2-RESEARCH.md.

Official-doc findings to reflect in Prompt Library import/indexing:

• gpt-image-2 is a direct Image API model, while this app mainly uses mainline gpt-5.x models with the image_generation tool through Responses API.
• Current size constraints match the app implementation: max edge 3840px, 16px increments, max 3:1 ratio, and 655,360 to 8,294,400 total pixels.
• quality should be indexed as low / medium / high / auto; low is useful for draft/thumbnail iteration.
• gpt-image-2 does not support transparent backgrounds, so imported prompts asking for transparency should get a warning/penalty for GPT Image 2 use.
• input_fidelity should be omitted for gpt-image-2; image inputs are always processed at high fidelity, which affects edit/reference-image cost.
• Search/indexing should preserve hints like gpt-image-2, image_generation, reference-image, mask-edit, typography, layout, 2k, 4k, low, medium, and high.

This does not expand PR1. PR1 remains import dialog/dropzone + local preview + single GitHub .md / .markdown / .txt preview/import. The GPT Image 2 metadata/search work belongs in PR2 indexing.