AI Skill Report Card
Curating Beading Pattern Indexes
Quick Start14 / 15
Given a raw URL list, produce output like:
Recommendation▾
Add an example showing a non-pattern/utility URL being correctly discarded, since this is a stated pitfall but not demonstrated
Dragons & Fantasy Creatures
Drachen (Dragon) — Series
- perlentiere.com "drachen" series — vorlagen/, German, PDF
- Variants 01–16: direct PDF, http://www.perlentiere.com/03_anleitungen/anleitungen/vorlagen/drachen{NN}.pdf
- Variants 20, 21, 25, 26, 28–30, 36–38: gallery-only, https://perlentiere.com/index.php?...&motiv=drachen{NN}
- Total: 24 distinct designs found
- Note: numbering is non-contiguous; gaps likely exist in unlisted range
S-Dragon
- S-Dragon Tutorial — DeviantArt (rrkra), full written tutorial
- http://rrkra.deviantart.com/art/S-Dragon-Tutorial-263218162
- Language: English, Format: webpage (step-by-step with images)
No preamble — start straight into classification and grouping.
Workflow14 / 15
Progress:
- Step 1: Ingest full URL list; classify each as PDF pattern, HTML/image diagram, blog tutorial, forum thread, gallery/index page, or non-pattern utility page
- Step 2: Discard or separately flag non-pattern URLs (homepages, FAQ, Impressum/contact, search-by-letter pages, social profiles)
- Step 3: Extract subject per URL using naming heuristics (see below); keep native term + English gloss
- Step 4: Deduplicate — merge same-subject entries across domains, CMS versions, and gallery-vs-PDF pairs
- Step 5: Collapse numbered/dated series into single entries with variant counts, not one bullet per URL
- Step 6: Assign each surviving entry to exactly one category (never duplicate across categories)
- Step 7: Annotate language, format, and series metadata for every entry
- Step 8: Render as markdown index following the fixed template; flag uncertainties inline
Recommendation▾
Consider showing a 'bad output' example (e.g., one-bullet-per-URL) alongside the correct dedup version to reinforce the contrast
Naming & Classification Heuristics
perlentiere.com/.../vorlagen/{subject}{NN}.pdf→ subject is German animal/object name. Translate for gloss (drachen=dragon, vogel=bird, schnecke=snail, eichhoernchen=squirrel, hase=rabbit, kuecken=chick) but keep the German slug for searchability.index.php?...&motiv={subject}{NN}→ gallery thumbnail, treat as "preview" alternate to the matching PDF if one exists; otherwise list as gallery-only.biserok.org/{russian-transliterated-slug}/→ extract subject from slug, gloss in English (bereza=birch, derevo=tree, bonsay=bonsai).- DeviantArt URLs → subject = artwork title segment before the ID number; artist = subdomain/username; treat as "primary tutorial" if it reads as a full stitch guide, "pattern only" if it's a single chart image.
beadjewelrymaking.com/{YEAR}/project/{month}.htmland similar monthly/dated project pages → treat as a chronological series, not individual top-level subjects, unless the subject is distinctive enough to warrant separate listing.perlentiere-forum.de/wp-content/uploads/{YEAR}/{MONTH}/{NN}_{Subject}.pdf→ belongs to a dated contest/advent-calendar series; preserve series name + date range + numbering in one series block.- Numbered variants of the same creature (e.g.,
drachen01–drachen38) → one entry, variant range + count, not 38 bullets.
Categories (use only these)
- Animals → subdivide inline as Mammals / Birds / Fish / Reptiles & Amphibians / Insects & Arachnids
- Dragons & Fantasy Creatures
- Pokémon & Pop-Culture Characters
- Plants & Trees
- Decorative Objects & Ornaments
- Techniques & Grundlagen
- Site Indexes & Collections (meta-resources, listed once per site, never per search-letter page)
Examples17 / 20
Example 1:
Input: drachen01.pdf through drachen16.pdf, plus gallery motiv=drachen36
Output:
### Drachen (Dragon) — Series
- Patterns 01–16: direct PDF, vorlagen/drachen{NN}.pdf
- Patterns 20, 21, 25, 26, 28–30, 36–38: gallery-only, motiv=drachen{NN}
- Total: 24 known variants (gaps 17–19, 22–24, 27, 31–35 unaccounted)
Example 2:
Input: http://rrkra.deviantart.com/art/Axolotl-tutorial-580727616
Output:
Recommendation▾
The skill is highly niche (specific to perlentiere.com, biserok.org, DeviantArt) — a brief note on how to generalize heuristics to new/unknown domains would improve robustness
Animals > Reptiles & Amphibians
Axolotl
- Axolotl Tutorial — DeviantArt (rrkra), full written tutorial
- http://rrkra.deviantart.com/art/Axolotl-tutorial-580727616
- Language: English, Format: webpage (primary tutorial, stitch-by-stitch)
**Example 3:**
Input: Two URLs for the same creature, old (`www.perlentiere.com`) and relaunch (`perlentiere.com`) paths
Output: Single merged entry — "mirrored/legacy path — verify which is live before linking" with both URLs listed, PDF as primary.
Best Practices
- Always keep the original URL verbatim, even if dead — enables archive.org lookup.
- Flag file format explicitly (.pdf / .htm / .gif / .jpg / interactive gallery) so users know what tool they need.
- Give English gloss alongside native term for all non-English subjects — never drop the original slug.
- Prefer PDF as primary link; gallery/thumbnail page listed as secondary "preview."
- Keep dated/numbered series intact and chronologically or numerically ordered rather than scattering variants through the index.
- Note domain relationships explicitly (e.g., "perlentiere.com vs. perlentiere-forum.de — related but distinct platforms, do not silently merge content").
Common Pitfalls
- Don't create one bullet per URL — this defeats the purpose of the index; always dedupe by subject first.
- Don't treat utility pages (FAQ, Impressum, contact, social profiles, search-by-letter) as patterns.
- Don't group by domain — a single domain hosts dozens of unrelated subjects; always regroup by subject.
- Don't assume shared domain root means same ownership/content set — verify before merging.
- Don't translate away the original term — "Pummeluff" must stay next to "Jigglypuff," not replace it.
- Don't flatten numbered variant series into a single generic bullet — report the variant count explicitly.