菜单
Read, create, and modify Apple Keynote (.key) presentations. Use when the user asks about .key files, wants summaries, needs to understand slide content, or wants to create/edit decks.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | keynote |
| description | Read, create, and modify Apple Keynote (.key) presentations. Use when the user asks about .key files, wants summaries, needs to understand slide content, or wants to create/edit decks. |
| user_invocable | true |
Work with Apple Keynote (.key) presentations: extract text, export images, create new decks, add slides, set text, populate entire decks from JSON, and insert images (incl. assets fetched from the web).
$ARGUMENTS is a file path, glob pattern, or instruction. If empty, default to all .key files in the current working directory.
.key file)extract-text — get slide text, titles, and presenter notes:
osascript .claude/skills/keynote/scripts/extract-text.applescript "<absolute-path>"
export-images — export slides as PNGs:
osascript .claude/skills/keynote/scripts/export-images.applescript "<absolute-deck-path>" "<absolute-output-dir>"
Both paths must be absolute — a relative output dir (e.g. ./.keynote-skill/images/<name>) fails with "could not be interpreted as a file URL". Use "$(pwd)/.keynote-skill/images/<name>".
.key file — automatic backups)All write scripts back up the deck to <deck-dir>/.keynote-skill/backups/<deck-name>.<timestamp>.key before modifying, then prune so only the 3 most recent backups per deck remain. The deck itself can live anywhere — you can collaborate directly on the user's working file.
All scripts (read-only and write) leave the deck open in Keynote by default so the user sees results immediately and isn't disrupted if they were already working on it. Pass --close as the last argument to close the deck after the operation (useful for batch scripting or when you want a clean state).
Auto-save before edit: every script first calls saveIfOpen — if Keynote currently has the target deck open, the script saves it before proceeding. This commits any unsaved user edits to disk before the backup runs, so the backup captures their latest work and the script's own save doesn't clobber pending changes. No need to ask the user to save first.
create-deck — create a new Keynote file using the talk-basic theme. Refuses to overwrite an existing file:
osascript .claude/skills/keynote/scripts/create-deck.applescript "<absolute-path>"
delete-slides — remove slides by number/range:
osascript .claude/skills/keynote/scripts/delete-slides.applescript "<deck-path>" "<slide-spec>"
<slide-spec> accepts comma-separated numbers and ranges: "3,5,10-15,20"
add-slides — append slides with a specific master layout:
osascript .claude/skills/keynote/scripts/add-slides.applescript "<deck-path>" "<master-name>" <count>
set-slide-text — set title, subtitle, and body on a single slide:
osascript .claude/skills/keynote/scripts/set-slide-text.applescript "<deck-path>" <slide-number> "<title>" "<subtitle>" "<body>"
Pass "" for unused fields. For Title layout, <body> sets the footer/author field. For Title Only layout, <body> creates a freeform text box below the subtitle.
populate-deck — populate an entire deck from a JSON file. Destructive: refuses to run on a deck with >1 slide unless --force is passed (the user's edits would be clobbered):
osascript .claude/skills/keynote/scripts/populate-deck.applescript "<deck-path>" "<slides.json>" [--force]
Slide 1 is reused for the first JSON entry; additional slides are appended. Processes in batches of 5 to avoid timeouts. Prefer set-slide-text + add-slides for incremental edits — only use populate-deck to build a deck from scratch.
insert-image — insert an image (PNG/JPEG/etc) onto a slide:
osascript .claude/skills/keynote/scripts/insert-image.applescript "<deck-path>" <slide-number> "<image-path>" "<position>" [width]
<position> accepts named keywords (bottom-right, bottom-left, top-right, top-left, center) or explicit "x,y" (canvas is 1920×1080). [width] defaults to 500; height auto-scales by aspect ratio.
add-text-box — add a freeform text box anywhere on a slide (custom position, font, size):
osascript .claude/skills/keynote/scripts/add-text-box.applescript "<deck-path>" <slide-number> "<text>" "<position>" [width] [font] [size]
Use this for any standalone text box that isn't one of the layout's title/subtitle/body placeholders — e.g. labels, data cards, callouts, replacing a deleted image with text. <position> accepts the same named keywords as insert-image or explicit "x,y". Use \n (literal backslash-n) in <text> to separate lines — each becomes its own line in one box. [width] defaults to 560, [size] to 32; [font] (PostScript name like Menlo-Bold) is optional — omit to keep the theme default. Prefer this over hand-rolling AppleScript — it bakes in the one creation pattern that actually works (see pitfalls: text items must be made inside a tell <slide> block with a height property, or you get -10000).
When asked to add a logo, screenshot, or other image to a slide:
Find a source — good defaults:
WebSearch then WebFetch to get the direct asset URL.Download to assets dir:
mkdir -p ./.keynote-skill/assets
curl -sL -A "Mozilla/5.0" "<asset-url>" -o "./.keynote-skill/assets/<name>.png"
The User-Agent header matters — Wikimedia and others 403 default curl UAs.
Verify with file ./.keynote-skill/assets/<name>.png — should report PNG image data (not HTML document — a sign the URL returned an error page).
Insert via the insert-image script.
Tips:
commons/X/XX/Foo.svg with commons/thumb/X/XX/Foo.svg/<width>px-Foo.svg.png.<name>-color.png (e.g. claude-color, gemini-color, antigravity-color). For brands whose default icon is already colored (e.g. openai), the base name works. Try the -color variant first; fall back to the monochrome only if no color variant exists.curl and Mozilla/5.0 UAs get HTTP 400. Use a UA with contact info, e.g. curl -A "LogoFetcher/1.0 (shawhintalebi@gmail.com)" ....https://unpkg.com/@lobehub/icons-static-png@latest/light/<name>.png. Append -color for the color variant. Doesn't have python/javascript/ruby/java/cpp/html/word/excel/etc — those come from Wikimedia./Applications/ first. When you want the icon "as shown on the desktop app" (purple-gradient Codex, Claude, etc.), the real icon lives at /Applications/<App>.app/Contents/Resources/icon.icns (sometimes AppIcon.icns). Convert with sips -s format png <path>.icns --out ./.keynote-skill/assets/<name>-app-icon.png. This is faster than web search and returns the actual shipping icon — branded marketing logos on the web are often different from the app icon (e.g. Codex ships a black wordmark logo and a purple gradient app icon; only the latter is in Codex.app).Keynote's Movie inspector exposes only volume, trim, poster frame, and repeat — there's no playback-speed slider. To present a screen recording faster, pre-process it:
.claude/skills/keynote/scripts/speed-up-video.sh <input> <speed> [output]
Wraps ffmpeg with setpts for video and chained atempo filters for audio (preserves pitch at any speed). Default output is <input-stem>.<speed>x.<ext> next to the input. Replace the original movie in the slide via the Movie inspector's Replace button.
| Layout | text item 1 | text item 2 | text item 3+ |
|---|---|---|---|
| Title | footer/author (y=934) | big title (y=203) | subtitle (y=569) |
| Section | center text (y=357) | — | — |
| Title Only | title (y=85) | subtitle (y=187) | — (no body placeholder) |
| Blank | — | — | — |
The scripts detect which layout a slide uses by checking the y-position of text item 1:
> 900 → Title layout> 300 → Section layout< 200 → Title Only layoutEach layout has duplicate/extra text items (items 4+) that are ignored. Item at y=1030 is a logo/watermark placeholder.
Body text on Title Only slides: Since there is no body placeholder, body content is injected via make new text item as a freeform text box positioned at y=300. These are starting points for manual editing.
Each newline in the body field becomes a separate bullet point. Use \n (literal newline) to separate bullets in JSON:
{"body": "First bullet\nSecond bullet\nThird bullet"}
For a subtitle line without a bullet (on Title Only), use the subtitle field, not the body field.
[
{"master": "Title", "title": "Deck Title", "subtitle": "A subtitle", "body": "", "author": "Your Name"},
{"master": "Title Only", "title": "Slide Title", "subtitle": "Lead-in line", "body": "Bullet 1\nBullet 2\nBullet 3"},
{"master": "Blank"}
]
Title, Section, Title Only, Blank
slides.json file following the format aboveosascript .../create-deck.applescript "<deck-path>"osascript .../populate-deck.applescript "<deck-path>" "<slides.json>"set-slide-text / add-slides / delete-slides / insert-image to make targeted edits. Do not re-run populate-deck on an in-progress deck — it refuses by default and even --force discards user edits.<deck-dir>/.keynote-skill/backups/ before modifying. Read-only scripts (extract-text, export-images) make no backup.create-deck refuses to overwrite an existing file.populate-deck refuses to run on a deck with >1 slide without --force. This is the only destructive script — when in doubt, use the additive scripts instead.<deck-dir>/.keynote-skill/backups/ (auto-created by write scripts)./.keynote-skill/assets/./.keynote-skill/images/<presentation-name>/./.keynote-skill/Run mkdir -p ./.keynote-skill/images ./.keynote-skill/assets before writing outputs.
When creating new presentations, first read .claude/skills/keynote/style-guide.md. Follow the visual design, narrative structure, and content patterns described there to match your established presentation style.
Common boilerplate (path resolution, --close parsing, save-if-open, backup) lives in scripts/_helpers.applescript. Each script loads the compiled version (_helpers.scpt) at runtime via load script, since AppleScript's load script only accepts compiled .scpt, not raw source.
If you edit _helpers.applescript, recompile it:
osacompile -o scripts/_helpers.scpt scripts/_helpers.applescript
Both files should be checked in. The .applescript is the source of truth for reading/editing; the .scpt is what the scripts actually load.
Exported handlers:
resolvePath(p) — relative or absolute path → fully resolved absolute pathparseCloseFlag(argv) — true if argv contains --closesaveIfOpen(filePath) — saves the deck if Keynote currently has it open (commits user's unsaved edits before backup/edit)backupDeck(filePath) — copies the deck to <deck-dir>/.keynote-skill/backups/<name>.<timestamp>.key, then prunes so only the 3 most recent backups for that deck remainslide N of doc immediately after creation in the same tell block. Use direct references from make new slide or operate in a separate open/save/close cycle.populate-deck script handles this automatically.width (and position/file) as creation properties to make new image. Setting width or height after the image is created has been observed to corrupt the .key file. The insert-image script handles this correctly.-10000 creating a text box: A new text item must be created inside a tell <slide> block with make new text item with properties {object text:…, position:…, width:…, height:…} — the height property is required. Using make new text item at end of slide, or omitting height, throws AppleEvent handler failed (-10000). This is a creation-structure problem, not a special-character/Unicode issue — don't waste time stripping °/arrows/etc. Set font/size on object text after creation. The add-text-box script encapsulates the working pattern, so reach for it instead of hand-rolling.text items shadowing inside Keynote tell blocks: Inside tell application "Keynote", the term text items refers to slide text-item objects, not string parts. If you parse comma-separated args (e.g. "180,550") by setting AppleScript's text item delimiters and reading text items of <string>, do it outside the Keynote tell block — otherwise you'll get error -1728 ("Can't get every text item of '180,550'"). Same applies to any keyword that overlaps Keynote's vocabulary.populate-deck will wipe their changes. Use set-slide-text for text fixes and insert-image for adding visuals.populate-deck script requires Python 3 (for JSON parsing)AI Builder Academy's brand and design system — colors, typography, voice/tone, components, icons, and copy patterns for aibuilder.academy and any ABA-branded surface (landing pages, decks, one-pagers, social graphics, emails, motion graphics). Use whenever Shaw is designing, mocking up, or reviewing anything that should look or sound like ABA. Triggers: "design a landing page for ABA", "make an ABA mockup", "one-pager for [ABA offer]", "what colors / font does ABA use", "ABA-ify this", "make this on brand", "what's the ABA voice", "design an ABA banner / hero / card / form", "slide template for ABA", "match the ABA homepage", "ABA infographic", "ABA animation". Also fire on any AI Builder Academy reference when the deliverable is a visual artifact (HTML, slide, image, mockup, SVG, React component, MP4, PNG) or copy that sits on an ABA surface. Casual mentions like "make it look like the ABA site" or "give this an aibuilder.academy feel" count too.
Shaw's strategic thinking partner for AI Builder Academy business decisions. Use this skill whenever Shaw is thinking through business strategy, customer avatars, market positioning, pricing, go-to-market plans, outreach strategies, offers, funnels, or any high-level business decision. Triggers include: "let's think through," "help me figure out," "strategy session," "who should I target," "what should I charge," references to avatars, ICPs, customer segments, offers, funnels, or any open-ended business planning conversation. Also trigger when Shaw is clearly brainstorming — talking through ideas conversationally, expecting Claude to be a concise sparring partner that grounds decisions in real data, asks sharp questions, and pushes back honestly. Even casual phrasing like "I've been thinking about," "what do you think of," or "help me decide" should trigger this skill when the topic is business-related.
Create, reschedule, and delete events on Shaw's Google Calendar — and keep the matching Notion page in sync when one exists. Use this skill whenever Shaw wants to put something on the calendar, move an existing event, or cancel one. Triggers include: "schedule X", "create an event for", "put X on my calendar", "book a call with Y next Tuesday", "set up a 1:1 with", "another session with X next Saturday", "follow-up call with X", "reschedule the X meeting", "move my X call to Friday", "cancel the X event", "delete that meeting", or any casual phrasing that implies adding/changing a calendar block (e.g., "let's do another one with [Client Name]", "block off Wednesday morning for the workshop"). Even when Shaw doesn't say "calendar" explicitly, this skill should fire if the request is about scheduling. This skill also creates a paired Notion page in ABA Calls or ABA Trainings when the event is a sales call or a delivery session and a paired page is part of the existing pattern.
Import or update skills in this public repo from Shaw's canonical skills repo at ~/agents/skills. Use whenever Shaw says "sync skills", "update the non-coding skills", "pull <skill> from canonical", "add these skills to the repo", or otherwise references moving his private canonical skills into this public repo. Handles PII scrubbing via parallel sub-agents, README table maintenance, and produces a clean commit. Do NOT use for creating skills from scratch (→ skill-creator) or one-off edits to a single already-imported skill.
Draft, sharpen, or diagnose copy whose job is conversion — landing pages, event registration pages (Luma), one-pagers, LinkedIn posts meant to sell, newsletters meant to sell, sales emails, and cold outreach DMs. Use whenever Shaw asks to write, rewrite, review, tighten, or critique conversion-focused copy. Triggers include "write the landing page", "draft the one-pager", "sharpen this headline", "tighten this newsletter", "rewrite this CTA", "write me a cold email", "outreach DM", "review this copy", "make this hit harder", "sounds like a brand wrote it". Also trigger when Shaw shares a copy draft and asks for review or rewrite, or casual phrasing like "help me write the promo". Applies Harry Dry's 3-question filter plus universal and channel-specific frameworks. Do NOT use for general LinkedIn posts (→ linkedin-post-writer), top-of-funnel essays (→ linkedin-post-writer or newsletter-writer), full sales letter structure (→ sales-letter-writer), or emails without conversion intent (→ email-helper).
Use this skill whenever Shaw mentions leads, contacts, sales pipeline activity, or anything CRM-related — even casually. Triggers include: "check my CRM", "update the CRM", "any updates on leads?", "add this person to the CRM", "review my pipeline", "follow up with X", "anyone worth following up with", "check active campaigns", "did anyone respond?", "cross-reference Gmail", "ABA contact form", or any mention of a specific lead's name in a sales context. Also triggers when Shaw shares new contact form submissions or inbound emails that look like potential leads. Also triggers for client/nurture activity: "check on my clients", "any clients due for check-ins", "move this lead to clients", "nurture review", or any mention of existing clients and expansion opportunities. Also triggers when a discovery call is booked — directly or surfaced during a sync. Follow-up reviews cover Active Leads, Clients (Nurture), AND any active campaign trackers linked from the CRM page.