日本語

slide-forge Settings

Two switches live in config/settings.json and apply to every run, so they are settled here once instead of being asked at every intake. The reference is references/settings.md; this skill is the dialogue around it.

Important

• Never touch anything else in config/. credentials.json, token.json, and gemini_api_key are secrets; this skill only reads and writes settings.json, and never prints a key or token.
• Show before you change, and read back after. Every run starts with --show and ends with --show, so the user sees what changed.
• Don't ask what the user already said. "画像生成をオフにして" is a complete instruction — apply it, report it, and only ask about the other switch if they asked to review settings generally.
• A change never touches an existing deck. It affects the next generation. Say so when the user is mid-workflow.
• Run every command from the slide-forge root as cwd — ${CLAUDE_PLUGIN_ROOT} when running from an installed plugin, /path/to/slide-forge on a local clone.

Quick Reference

Flags combine in one call. Precedence, weakest first: defaults → config/settings.json → GSLIDES_IMAGE_GENERATION / GSLIDES_OUTPUT / GSLIDES_LOCAL_DIR → build_deck.py --output. If --show reports an environment override, say so — the file the user is editing is not what that run will use.

Workflow

1. Show the current values, and quote them in the question descriptions so the user is choosing against a known state:

   .venv/bin/python scripts/settings.py --show

2. Ask with AskUserQuestion — one round, both switches, current value marked. Ask only about switches the user has not already decided:

   {
     "questions": [
       {
         "header": "画像生成",
         "question": "Gemini による画像生成を使いますか？",
         "multiSelect": false,
         "options": [
           {"label": "使う（現在の設定）", "description": "aiImage 図・images.py・image-slots が使える。課金済みの GEMINI_API_KEY が要る（画像モデルは無料枠クォータが 0）"},
           {"label": "使わない", "description": "AI 画像を一切生成しない。aiImage は検証時に弾かれ、代わりに図形で描く illustrations / patterns を使う。API キーは不要"}
         ]
       },
       {
         "header": "出力先",
         "question": "成果物をどこに出しますか？",
         "multiSelect": false,
         "options": [
           {"label": "Google Drive / Google Slides（現在の設定）", "description": "生成したデッキの URL が成果物。共同編集・コメントができる。PPTX が要るときは pptx-export で個別に書き出す"},
           {"label": "ローカルフォルダ / PowerPoint", "description": "生成のたびに out/pptx へ .pptx を書き出す。デッキ自体は編集可能な原本として Drive に残る（削除はしない）"}
         ]
       }
     ]
   }

   Put the current value first and label it 現在の設定, so the safe answer is the top option. When "ローカルフォルダ" is chosen and the folder should not be out/pptx, ask the path in a follow-up question — the free-text "Other" answer is the path.

3. Apply, combining every change into one call:

   .venv/bin/python scripts/settings.py --image-generation off --output local

4. Read back and report — the file path, the resulting values, and what changes in practice. Two consequences are worth stating explicitly:

   • image generation off → propose illustrations / patterns / diagrams for visuals from now on; specs carrying aiImage will fail offline validation until they are rewritten;
   • output local → the PPTX question disappears from intake, generation prints a local path next to the deck URL, and visual QA still runs against the Slides deck (export is a snapshot: QA first, then export).

When another skill sends the user here

The generation skills read these settings at intake (references/interactive-intake.md) and drop the questions the settings already answer. When a user asks to change one mid-workflow, run this skill, then continue the interrupted workflow with the new values — no regeneration is needed unless the deck was already built.