palette

name	palette
description	Create, revise, and render slide decks (PowerPoint .pptx) from a natural-language request or pasted material. Use whenever the user wants to build a presentation / deck / slides, turn notes or content into slides, or change a deck. Works in two reviewable steps — a markdown plan the user approves, then a rendered .pptx.

Palette — slide deck builder

Palette turns a request (or pasted material) into a rendered slide deck in two reviewable steps:

1. Plan — a markdown outline of the deck, which the user reviews and approves.
2. Deck — a rendered PowerPoint .pptx, built from the approved plan.

Always let the user approve the plan before building the deck. The plan is the cheap place to iterate; the deck build renders every slide and takes a few minutes, so you don't want to build from an unapproved plan.

Commands

Run these from the Palette harness directory (where palette.py lives), with the harness environment configured (model access via RITS_API_KEY). Each command prints its result to stdout and, on failure, prints error: ... and exits non-zero — relay that message to the user.

1. Create a plan

python palette.py build-plan "<the user's request>" --out plan.md

• Pass the user's request through as-is — do not reformat or restructure it.
• If the user pasted material for the deck (notes, content, data, an excerpt), add it with --context so the plan is grounded in it:

  python palette.py build-plan "<request>" --context "<the pasted material>" --out plan.md
  

• Output: the markdown plan (also saved to plan.md).

2. Revise a plan

python palette.py edit-plan "<the change the user asked for>" --plan plan.md --out plan.md

• Use for any change to the plan: slide count, tone, wording, content, adding or removing a slide, changing a specific slide.
• Pass the user's requested change through as-is.
• Output: the full revised plan.

3. Build the deck (renders the .pptx)

python palette.py build-deck --plan plan.md --out-dir <output_directory>

• Runs the full pipeline and writes deck.pptx (plus deck.json and slide preview PNGs) into <output_directory>.
• Output: the path to the rendered .pptx on stdout.
• This is slow (~minutes) — tell the user it's building before you run it.
• Optional: --palette-family <style> sets the visual style (default ibm_watsonx). Only use it if the user asks for a specific look.

Workflow

The loop is build-plan → confirm → [edit-plan → confirm] × N → build-deck.

1. User asks for a deck → run build-plan with their request (add --context if they pasted material). Save the plan.

2. Confirmation gate (required). Present the plan to the user and explicitly ask them to confirm before anything is built — for example:

   Here is the plan for your deck. Does this look right? Reply yes to build the deck, or tell me what you'd like to change.

   Never call build-deck until the user has confirmed this plan.

3. Branch on the user's reply:

   • Confirms ("yes", "looks good", "go ahead", "build it") → run build-deck, then give the user the .pptx path it returned.
   • Asks for changes ("no…", "change…", "add a slide about…") → run edit-plan with their instruction, then return to step 2 — present the revised plan and ask for confirmation again.

4. Repeat until the user confirms, then build.

Guidance

• Keep your job simple. Pass the user's words straight through to build-plan / edit-plan. The tools do the parsing, grounding, and formatting — don't pre-process the request yourself.
• The confirmation gate is mandatory. Always present the plan and get the user's explicit confirmation before running build-deck — never build an unconfirmed plan, and never jump straight from the request to build-deck.
• Presentation context, not upload. This host may not support file uploads. If the user pastes their material into the chat, pass it via --context — it is used as the source material the deck is built from (shaped into slides, not copied verbatim).
• Report the path. After build-deck, the user finds their deck at the .pptx path the command printed. State that path clearly.
• For programmatic parsing, add --json to any command to get a JSON envelope ({"ok": true, ...}) instead of raw text.