메뉴
Read, extract from, combine, split, watermark, fill, or produce PDF files. Use whenever a .pdf is the input or the deliverable — merging reports, pulling text or tables out, rotating scans, filling forms, or generating a polished PDF from content.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Produce or modify Word documents (.docx) as real files — reports, memos, letters, proposals, or any deliverable the operator will open in Word. Use when the requested output is a Word file, or when an existing .docx needs content extracted, edited, or restructured. Not for spreadsheets, decks, or PDFs — those have their own skills.
Build or edit slide decks (.pptx) as real files — pitch decks, briefings, status readouts — or extract content from existing ones. Use whenever slides or presentations are the input or the deliverable. Not for documents, spreadsheets, or PDFs.
Produce, edit, or clean spreadsheets (.xlsx/.csv/.tsv) as real files — reconciliations, trackers, budgets, data cleanups, format conversions. Use whenever the deliverable is a spreadsheet or an existing one needs work, including messy tabular data that needs restructuring. Not for Word/PDF/deck outputs.
Compose the operator's daily brief — schedule, things waiting on them, ongoing work, and heads-ups — rendered as a polished HTML artifact, optionally on a recurring schedule. Run only when the operator explicitly asks for their brief (or /daily-brief), or when executing a scheduled brief task. An ordinary question about today's calendar is answered directly, not with a brief.
Conversational first-run companion for the Cowork archetype — verify the workspace is ready, import existing Claude Code/Cowork state, prove the setup on a real task, learn the operator's writing voice, and wire the first scheduled task. The mechanical setup (archetype, plugins, dependencies, folders) belongs to the setup wizard; this skill picks up where it leaves off.
Reflective maintenance pass over long-term memory — merge duplicates, retire stale or completed items, sharpen what stays. Use when the operator asks to clean up/organize memory, when recall keeps surfacing outdated facts, or periodically as a scheduled task.
| name | |
| description | Read, extract from, combine, split, watermark, fill, or produce PDF files. Use whenever a .pdf is the input or the deliverable — merging reports, pulling text or tables out, rotating scans, filling forms, or generating a polished PDF from content. |
Two different jobs, two libraries — both via execute_code:
from pypdf import PdfReader, PdfWriter
# extract text
text = "\n".join(page.extract_text() for page in PdfReader(src).pages)
# merge
w = PdfWriter()
for part in [a, b, c]:
w.append(part)
w.write(out)
# split / rotate / subset
w = PdfWriter()
for page in PdfReader(src).pages[2:5]:
page.rotate(90)
w.add_page(page)
w.write(out)
Also: form fill (w.update_page_form_field_values), encrypt/decrypt
(w.encrypt(pwd) / PdfReader(src, password=pwd)), metadata
(reader.metadata), attachments and watermarks (page.merge_page(stamp)).
Scanned PDFs: extract_text() returning almost nothing means image-only
pages. Say so plainly and offer what's available (OCR only if an OCR tool is
actually installed — check, don't assume).
For a generated deliverable (report, one-pager, cover sheet):
from reportlab.lib.pagesizes import LETTER
from reportlab.lib.styles import getSampleStyleSheet
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer
styles = getSampleStyleSheet()
doc = SimpleDocTemplate(path, pagesize=LETTER)
doc.build(
[Paragraph("Title", styles["Title"]), Spacer(1, 12)]
+ [Paragraph(p, styles["BodyText"]) for p in paragraphs]
)
If the content is really a Word document the operator also wants as PDF, build the .docx (docx skill) and produce the PDF alongside it — don't force page layout through reportlab when Word styling is the source of truth.
Re-open the output: page count, and text extraction on a sample page for generated PDFs. Report the verified page count with the saved path.
Once the file is written and verified, register it in the Artifact panel with
save_file_artifact(path, title=…) when that tool is available (the artifact plugin, protoAgent
v0.107.0+). It stores the bytes, shows a Download card with a readable preview (the extracted text), and keeps
an edit history — pass the same artifact_id to save a later revision as v2/v3. Report the saved
path and that it's in the panel. If the tool isn't available, just report the path.