Organize docs from zip (magj.dev)

When this applies

The user supplies a local zip path (for example ./samples.file.zip). The agent unpacks safely, classifies each Markdown file, writes or proposes paths under src/content/blog/ or src/content/references/, aligns tone with the site, and does not treat the result as published until a human confirms.

This is a local editorial workflow. It complements (does not replace) the admin References zip pipeline described in reference.md.

Quick checklist

- [ ] Confirm zip path; refuse if missing or not a file
- [ ] Inspect archive (list entries; no path traversal)
- [ ] Extract to a disposable staging directory under the repo or /tmp
- [ ] Parse each .md: frontmatter + body; classify blog vs references
- [ ] Normalize frontmatter to project Zod schemas (see reference.md)
- [ ] Choose filenames (kebab-case slugs); avoid overwriting without explicit user OK
- [ ] Edit prose for magj.dev tone (see Tone)
- [ ] Set draft for human review (blog defaults draft; force draft on references until publish)
- [ ] Leave REVIEW.md or inline PR notes: sources, assumptions, open questions

Safe unpack

1. List only first: unzip -l <zip> (or equivalent) to see paths and sizes.
2. Reject entries containing .., absolute paths, or non-UTF-8 names you cannot normalize.
3. Extract to something like .staging/docs-import-<date>/ (gitignored if needed) or /tmp/magj-docs-import/.
4. Respect rough limits aligned with production zip handling: on the order of tens of Markdown files and tens of MB uncompressed; if larger, stop and ask to split the archive.

Do not execute binaries from the zip. Only treat .md (and optionally .mdx if the project supports it—this repo uses .md in content loaders) as first-class import targets.

Classify: blog vs references

If the active content.config.ts differs from reference.md, the repo file wins.

Tone (magj.dev)

• Clear, technical, complete sentences—not telegraphic bullet chains as the only prose.
• Prefer short paragraphs; explain decisions where useful.
• Match existing posts: read one or two current files from src/content/blog/ and src/content/references/ before heavy rewriting.
• For deep edits, follow edit-article and blog-post-with-diagrams when diagrams help.

Frontmatter and drafts

• Blog (src/content/blog/): title, summary, publishedAt required per schema; keep draft: true until the human approves.
• References (src/content/references/): title, description, date required; set draft: true for this workflow even though the schema default may be false—human toggles off when publishing.

Normalize dates to YYYY-MM-DD. Derive missing titles from the first H1 or filename. Add sensible tags from content.

Slugs and collisions

• Filename = kebab-case slug + .md.
• If a path already exists, do not overwrite: use a suffix (-import, -2) or stop and list options for the user.

Handoff for human review

Deliver:

1. Table of files: source path in zip → repo path → collection → draft status.
2. Assumptions: anything inferred (dates, tags, blog vs references).
3. Checks: suggest bun run lint and bun run build (or project-standard commands) after edits.
4. Do not merge or deploy unless the user explicitly asks.

Additional detail

• Schema tables and zip pipeline alignment: reference.md