| name | adhoc_pdf_skills |
| description | Use when tasks involve reading, creating, or reviewing PDF files in this vault. Prefer rendered-page checks when layout matters. |
Ad-hoc PDF Skills
When to use
- Read or review PDF content where layout and visuals matter.
- Create PDFs programmatically with reliable formatting.
- Extract text or tables from PDFs for analysis.
Workflow
- Prefer visual review by rendering pages to images when possible.
- Use
reportlab for newly generated PDFs.
- Use
pdfplumber or pypdf for extraction and quick checks.
- For a new PDF, write the artifact only under
70_Exports\YYYY\MM\DD\<file_name>.pdf.
- For an edit request, read the original file, create a copy under
70_Exports\YYYY\MM\DD\<original-file-name>.pdf, and update only that exported copy.
- Re-render after meaningful updates and verify spacing, alignment, and legibility.
- In the final response, state that the source file was left untouched and provide the exported path for manual pickup.
Repo conventions
- Write all new or updated artifacts only under
70_Exports\YYYY\MM\DD\.
- Keep temporary render output under the same dated tree, for example
70_Exports\YYYY\MM\DD\tmp\pdf\.
- Never overwrite, rename, move, or save back to the original source path.
- Keep filenames stable and descriptive.
Dependencies
Install into the repo virtual environment with pip, not uv.
Primary install:
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
Targeted install:
.\.venv\Scripts\python.exe -m pip install reportlab pdfplumber pypdf
System tool for visual rendering:
If pdftoppm is missing, continue with extraction or generation work and note that visual QA could not be completed.
Rendering command
pdftoppm -png <exported.pdf> 70_Exports\YYYY\MM\DD\tmp\pdf\page
Quality checks
- Check headers, footers, pagination, tables, and image clarity.
- Avoid clipped text, overlapping elements, unreadable glyphs, and broken tables.
- Remove intermediate renders unless the user wants them kept.
