moonlitlynx-food-run

What This Skill Does

Finds media files that match digikam's "untagged or undated" saved search within the grouchygiraffe and shamblingshark collections, reads their YAML metadata, classifies food vs. non-food content, and applies the appropriate digikam tags.

Always report what you find before applying any tags.

Database

Path: ~/Pictures/digikam4.db

Back up before every run:

cp ~/Pictures/digikam4.db ~/Pictures/digikam4.db.bak.$(date +%Y%m%d_%H%M%S)

Finding the Files

The digikam saved search "untagged or undated" (id=42) translates to this SQL.

Both conditions must be true (AND, not OR):

SELECT i.name, ar.label
FROM Images i
JOIN Albums al ON al.id = i.album
JOIN AlbumRoots ar ON ar.id = al.albumRoot
LEFT JOIN ImageInformation ii ON ii.imageid = i.id
WHERE ar.label IN ('grouchygiraffe', 'shamblingshark')
AND i.id NOT IN (
  SELECT imageid FROM ImageTags
  WHERE tagid NOT IN (SELECT id FROM Tags WHERE name LIKE 'Color Label%')
)
AND (ii.creationDate IS NULL OR ii.creationDate < '1904-01-01T00:00:00')
ORDER BY ar.label, i.name;

The Color Label% exclusion is required — files with only a color label tag are treated as untagged by digikam's UI.

Reading YAML Files

For grouchygiraffe files: /Users/mtm/pdev/taylormonacelli/grouchygiraffe/data/<shortcode>.yaml

Strip the file extension to get the shortcode (e.g. DZY_GI9tCs5.mp4 → DZY_GI9tCs5.yaml).

Shamblingshark files in /output are UUID-named and have no YAML counterparts — skip classification, apply noyamldesc.

Read uploader, title, and description fields from each YAML.

Classification

Food — apply food (id=109) plus any matching subtags:

keyword pattern	subtag	digikam id
sourdough, levain, batard, boule, bulk ferment, crumb	sourdough	113
salsa, enchilada sauce, tomatillo, salsa verde, salsa roja	salsa	115
pepper, chile, jalapeño, chipotle, guajillo, cayenne	peppers	151
curry, masala, tikka, korma, biryani, desi, indian	indian	114
paneer	paneer	123
soup, stew, broth, ramen	soup	153
bean, lentil, chickpea	beans	133
tomato	tomato	131
coffee, espresso, barista	coffee	130
dumpling, potsticker, gyoza	dumpling	233
vegan, plant-based	vegan	234
pizza, pasta, focaccia, ciabatta, bread, loaf	pizza (pizza) or food only (bread)	235
mushroom	food only (no subtag exists)	—
honey	honey	156

Match against the lowercased description and uploader fields.

Apply food even when no subtag matches — food is the base tag for all food content.

Probably food / empty description — apply food + noyamldesc when the YAML description is empty or null but the uploader username strongly implies food content (e.g. eatinghealthytoday, thesourdoughnerd, katrina.the.earthy.vegan, myhealthydish).

No signal — apply only noyamldesc when description is empty and uploader gives no food signal (e.g. stealth_health_life with no description, generic lifestyle accounts).

Non-food — skip entirely. Do not apply any tags. Examples: fitness/mobility, fashion, political, medical, humor, lifestyle.

Special Tag: noyamldesc

noyamldesc (id=242, pid=0) marks files whose YAML has no useful description, making classification impossible.

Apply it alongside food when the username suggests food content, or alone when there is no content signal at all.

Always pair with ai-tagged.

Tag IDs Quick Reference

tag	id	pid
food	109	0
sourdough	113	109
salsa	115	109
indian	114	109
paneer	123	109
peppers	151	109
soup	153	109
tomato	131	109
beans	133	109
coffee	130	0
honey	156	109
dumpling	233	109
vegan	234	109
pizza	235	109
ai-tagged	192	0
noyamldesc	242	0

Applying Tags

Use INSERT OR IGNORE — the ImageTags table has a UNIQUE constraint on (imageid, tagid).

Look up image IDs by filename scoped to the correct album root:

SELECT i.id FROM Images i
JOIN Albums al ON al.id = i.album
JOIN AlbumRoots ar ON ar.id = al.albumRoot
WHERE ar.label = 'grouchygiraffe' AND i.name = 'DZY_GI9tCs5.mp4';

Or batch by name with subqueries in a single transaction:

BEGIN;
INSERT OR IGNORE INTO ImageTags (imageid, tagid)
SELECT i.id, 192 FROM Images i
JOIN Albums al ON al.id = i.album
JOIN AlbumRoots ar ON ar.id = al.albumRoot
WHERE ar.label IN ('grouchygiraffe', 'shamblingshark')
AND i.name IN ('DZY_GI9tCs5.mp4', ...);
COMMIT;

Scoping inserts by ar.label avoids tagging duplicate filename entries from other collections.

Workflow

1. Back up the database.
2. Run the untagged/undated query and report the file count.
3. Read YAML for each file and classify: food, probably-food, no-signal, or non-food.
4. Present findings to the user before applying any tags. Show: shortcode, content summary, proposed tags.
5. On confirmation, apply tags in a single transaction.
6. Verify counts by checking SELECT t.name, COUNT(*) FROM ImageTags it JOIN Tags t ON t.id = it.tagid JOIN Images i ON i.id = it.imageid WHERE i.name IN (...) GROUP BY t.name.
7. Update moonlitlynx progress.md in the Obsidian vault with the run summary.
8. Commit the progress note.

Progress Note

Progress is tracked in /Users/mtm/Documents/Obsidian Vault/moonlitlynx progress.md.

After each run, add a dated section under ## Food/recipe batch tag run documenting:

• File count found and tagged
• New tags created (if any)
• Breakdown of subtags applied
• Files skipped and why