Before executing the specific method, read Craft scenario commands and parameters, and execute according to the object selection, session, input replacement, and result checking within.
Task starting point
First read the materials, confirm the delivery purpose, target size or duration, modification scope, and required format. Details that can be determined from the task and materials should directly adopt reasonable defaults; only ask about missing conditions that would change the result.
When a visible editor is needed, hand off to the pdfcraft-editor skill. Standalone installation:npx skills add full-aigc-skills/pdfcraft-skills --skill pdfcraft-editor. The currently added skill is a local candidate; before release, use the local companion plugin, and do not treat this installation command as proof of release.
Execution steps
- First determine whether a text layer already exists, to avoid repeated OCR.
- Check the language, skew, resolution, and complex tables, and select one page for trial recognition.
- Perform recognition only when the current backend has the required language model.
- Spot-check names, numbers, and tables, and keep the original image and recognition layer.
Before recognition, confirm ocr_status.available and the actual language range; missing models and skipping pages that already have text do not count as actual successful scanned recognition.
Tool implementation
The upstream tool name belongs to PdfCraft. doc_open/doc_create return a doc handle; subsequent doc_info, doc_save, and page_render use the same doc. PDF page numbers start from 1, and parameters are subject to the actual schema.
First read the actual tool schema, then fill in the parameters; use the returned values for object or document IDs. Queried commands must be available in the current backend. Before saving, choose a new output path and keep the input. When the required algorithm, format, or command does not exist, explain the gap; do not pass off a similar name as the real thing.
Result checking
- Chinese capability must be actually available; English OCR must not be passed off as it.
- Text is searchable and reasonably positioned.
- Low-confidence content is marked for review; do not arbitrarily correct documentary facts.
Check the actual content against the user's goal, not just the process exit code. For local edits, inspect local details; for layout tasks, inspect the whole page; for time-based media, inspect the start, end, and key segments. For abnormal results, first check the current project to avoid repeatedly performing irreversible operations.
Delivery
Return the editable project, the actual derived files, and a brief explanation: what modifications were made, which steps were not executed due to current capabilities, and what content requires the user's judgment. List only files that actually exist and have been checked. If the user subsequently reworks it, read this source project and make local modifications, preserving the content that has already been accepted.