ワンクリックで
schematic-reader
Search schematic PDFs by text/OCR, then confirm details with region crops and OCR — not full-page re-reads.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Search schematic PDFs by text/OCR, then confirm details with region crops and OCR — not full-page re-reads.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Set health profile and daily calorie targets, log meals (text or photo), and suggest foods that fit remaining macros. Uses sylo_health_* tools; looks up restaurant items via web search (prefers sylo-web-access when enabled).
Personal trainer workflows — log workouts in natural language, look up lift history, plan weeks/months, analyze muscle gaps. Uses sylo_health_workout_* tools plus journal for pain/preferences.
Offline ME/CODESYS terminology, ST syntax, and bundled Schneider PDF manuals — read before writing PLC code or answering ME/M262 questions; do not search the web for basics covered here.
Edit Machine Expert file-based (.fbsproj) Structured Text on disk — no cds-text-sync or GitHub setup.
Read Excel (.xlsx) and LibreOffice (.ods) workbooks as structured JSON via read_spreadsheet.
When and how to use sylo_web_search and sylo_web_fetch for current/web information, and how to treat the untrusted content they return.
| name | schematic-reader |
| description | Search schematic PDFs by text/OCR, then confirm details with region crops and OCR — not full-page re-reads. |
| metadata | {"sylo":{"category":"documents","icon":"file-search"}} |
Use targeted search → region crop + OCR → brief answer. Full-page renders locate the sheet; region tools read wire numbers and terminal labels.
Industry pattern (engineering-drawing pipelines): detect/crop regions first, then run vision/OCR on the crop — full-page passes garble small text and invite hallucination.
| Tool | When |
|---|---|
search_schematic_pdf | Always first — locate pages; render_on_best_hit: true for a coarse full-page PNG |
render_schematic_page | Know the page; need layout context (one call only) |
render_schematic_region | Wire numbers / terminal labels too small on full page — default 300 DPI crop |
ocr_schematic_region | Deterministic text tokens near a terminal block; pass queries for wire/tag numbers |
["DGT-300+", "S-"]). Do not re-scan the whole drawing set.render_schematic_page (or render_on_best_hit) per question to find the area. If small text is unreadable, stop re-examining that PNG.render_schematic_region + ocr_schematic_region on the same bbox (normalized 0–1 fractions of page). Example bbox around lower-right detail: {"x0":0.55,"y0":0.35,"x1":0.98,"y1":0.85}.matched_tokens and region vision agree. If they disagree, widen bbox or raise dpi once — do not loop.search_schematic_pdf calls.Tag every wire/tag answer:
| Level | When to use |
|---|---|
| Verified (schematic) | OCR matched_tokens clearly show the wire at the terminal, or region crop vision unambiguously shows the label |
| Inferred (convention) | Color code, standard pinout, or industry practice — not read from the drawing |
| Unreadable | Region OCR empty and region vision ambiguous — say so; do not guess |
Never present Inferred as Verified. Example: "1242 (Verified — OCR on Actual Value S- terminal)" vs "likely 1242 (Inferred — WHITE is often S- on 4-wire cells; OCR did not confirm)".
query list is fine to prove the tool and build a sheet index.page_count and hit_count. hits is capped by max_results (default 8).bbox — {x0,y0,x1,y1} or [x0,y0,x1,y1] as fractions of page size (0–1). Use padding: 0.02 if labels clip.render_schematic_region / ocr_schematic_region — default 300 DPI on the crop.ocr_schematic_region.queries — e.g. ["1242", "S-", "1240"] to surface wire numbers in matched_tokens.use_ocr: true on search — scanned PDFs with empty text layer..schematic-reader/ beside the PDF.Wire to terminal: “What wire goes to S- on DGT-300+?”
{
"pdf_path": ".../drawing.pdf",
"query": ["DGT-300+", "S-"],
"render_on_best_hit": true
}
Then region on page from top hit:
{
"pdf_path": ".../drawing.pdf",
"page": 14,
"bbox": { "x0": 0.5, "y0": 0.4, "x1": 0.98, "y1": 0.88 },
"queries": ["1240", "1241", "1242", "1243", "S-"]
}
Run ocr_schematic_region with the same page/bbox/queries. Answer from matched_tokens when confident; otherwise Unreadable or Inferred with explicit label.
Survey: attached PDF → index queries, max_results: 12, no render until a detail question.