Combine skills

The four Office skills install side by side and are built to be combined: one request can read a spreadsheet with the Excel skill, pull narrative from a Word document, and deliver the result as a styled PDF or a presentation. This page shows how that works and how to prompt for it reliably.

How multi-skill tasks work

The agent orchestrates; the skills don't call each other. Each skill activates on its own trigger signals, and the agent loads as many as the task needs, running them in sequence and passing one skill's output files as the next one's input. Nothing special is required to enable this — a standard install already puts all four skills in the same folder, and their runtimes (one .NET, one Python, one Node.js) coexist by design.

Two consequences worth knowing:

• Work flows through files. A pipeline hop is always a file handoff (extracted text, an intermediate .md, a generated .xlsx), so give the agent a working folder and name the final output path.
• Steps run sequentially, which is what you want when outputs feed inputs — each stage can be verified with its own skill's acceptance checks before the next begins.

What flows into what

Plan a pipeline by matching one skill's outputs to the next skill's inputs:

The common bridge format is Markdown or plain text: every skill's content can be extracted to it, and the PDF skill's REFORMAT route consumes it directly. When a target skill doesn't read a source format (the PDF skill does not read .docx, for example), ask for an extraction step in between — the agent handles it as part of the pipeline.

Prompt recipes

PDF → editable Word document

Extract the text of vendor-agreement.pdf in this folder, then create
vendor-agreement-draft.docx from it - keep the heading and table
structure, standard contract formatting, and add a comment on every
clause that mentions payment terms so legal can review them.

The extraction hop pulls the PDF's text out first (the Word skill does not read .pdf), then the Word skill's CREATE pipeline rebuilds it as an editable .docx, adding the comments through its review workflow.

Spreadsheet analysis → presentation

Analyze sales-fy2026.xlsx in this folder: revenue per region per quarter
and the top three growth drivers. Then build sales-review.pptx from the
findings - 10 slides, corporate blue, one chart-style content slide per
region and a summary slide with the headline numbers.

The Excel skill does the analysis (never touching the source), then the PowerPoint skill turns the findings into a deck.

Word + Excel → styled PDF report

Extract the narrative from q3-summary.docx and the totals table from
q3-figures.xlsx in this folder, then produce q3-board-pack.pdf: report
type, deep navy accent, title "Q3 Board Pack" - narrative first, the
totals as a table block, and a callout with the single key takeaway.

Both sources are extracted first (Word and Excel skills), merged into content blocks, and typeset by the PDF skill's CREATE pipeline.

Spreadsheet → PDF form filling

Read the employee rows from new-hires.xlsx in this folder. For each row,
fill onboarding-template.pdf with the name, start date and department
fields, saving as filled/onboarding-<lastname>.pdf. Inspect the form
fields first and show me the mapping before filling.

The Excel skill supplies the data; the PDF skill's inspect-then-fill flow writes each form.

One source, two deliverables

From product-launch.md in this folder, produce both
launch-brief.pdf (proposal type, steel blue) and
launch-deck.pptx (12 slides, same accent color) - keep the
section structure identical across the two.

The same Markdown source feeds the PDF skill's REFORMAT route and the PowerPoint skill's CREATE route in one conversation.

Tips for reliable multi-skill prompts

• Spell out the pipeline when formats don't connect directly: "extract the docx content, then reformat it as a PDF" beats "turn this docx into a PDF" — the agent then plans the extraction hop instead of guessing.
• Name the intermediate and final paths. Explicit paths keep multi-step runs re-runnable and make each stage easy to inspect.
• Invoke skills by name when triggers collide. A prompt mentioning both "document" and "PDF" can route ambiguously — /minimax-docx and /minimax-pdf (Claude Code), the $ prefix (Codex), or "use the minimax-pdf skill to…" (CoDev/OpenCode) settles it.
• Verify at every stage, not just the end. Each skill has its own acceptance checks — the Excel formula validation, the PowerPoint markitdown re-extraction, the PDF make.sh check — and a multi-skill run is only as good as its weakest handoff. Ask for the per-stage evidence in the same prompt.