Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add wenhaochai/claude-plugins --skill talk-slidesgit clone --depth 1 https://github.com/wenhaochai/claude-pluginsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/wenhaochai/claude-plugins/talk-slides)<a href="https://agentmods.dev/skills/wenhaochai/claude-plugins/talk-slides"><img src="https://agentmods.dev/badge/skills/wenhaochai/claude-plugins/talk-slides/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/wenhaochai/claude-plugins/talk-slides"><img src="https://agentmods.dev/badge/skills/wenhaochai/claude-plugins/talk-slides.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00119 | $0.01567 |
| Opus 5.5 | $0.00048 | $0.00627 |
| Sonnet 5.5 | $0.00024 | $0.00313 |
| Haiku 4.5 | $0.00012 | $0.00157 |
Grade A, and why
talk-slides scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 51 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Talk slides
The Slides artifact type's own instructions cover the file format and the Artifact calls. This skill adds the author's rules, the layout builders and two checks.
Workflow
- Agree on the parts first, one idea per slide. Put open choices to the user as short multiple-choice questions, and cite only the sources the user names.
- Take every number from a primary source, keep the raw files next to the deck and recompute what you plot. Re-fetch live leaderboards on the day you publish and put the read date in the source line.
python3 scripts/build_template.py OUTwrites one example slide per layout: cover, part divider, figure, numbers, columns. Build slides withscripts/deckkit.py, draw charts per deck in its tokens, and replace every bracketed string.- Before each publish, run
python3 scripts/lint.py DECK_DIRandpython3 scripts/audit_render.py RUNTIME_DIR; the audit's docstring says how to set up RUNTIME_DIR. Then look at every slide at full size, a screenshot per slide, and fix each position conflict before the user sees the deck, without being asked: text that touches other text, a line or a mark; a leader line that crosses a label; arrowheads on labels; anything clipped or off the slide. A label sits closer to its own item than to the next one. Check each chart against its source. - Read the live deck before publishing so edits made in the page survive, then send only the changed files.
- Report in a few Chinese lines: the version, the link and what changed on which slide.
Slides
- Titles, slide and chart alike, are noun labels for what is shown and make no claim; none is a comma-joined list of clipped phrases. A title the user gives stays as given. The eyebrow names the part.
- The slide holds labels, numbers and at most one short sentence per block; the rest goes into the notes. Wording stays plain, without the "X, not Y" pattern, em dashes or the word campaign.
- The look is Google's, after the author's general exam deck: a white canvas with generous margins, and the four Google colors used to mark the parts of the talk and the one number that matters on a slide, never as decoration. Each slide's kicker carries a dot in its part's color; a part opens on a tonal divider. Nothing sits in a box: no tinted card tiles, no bordered panels; whitespace and hairlines separate the parts, big numbers stand on the white ground, and flows and comparisons are drawn as lines, timelines and tables.
deckkit.pyholds the tokens. - Type is restrained and the same on every slide. Google Sans sets the display type: the cover (104), part dividers (96), slide titles (52), big numbers (84), the one lead line of a slide (32, in grey), panel and chart headings (28, medium) and the kicker (24, medium). Google Sans Text sets the rest: 26 for body text and table cells, 24 for labels, captions and sources. Weights are regular and medium only; emphasis is the accent color, and nothing is italic. An animation's canvas text keeps the same sizes and weights. The lint flags type off this scale.
- Data goes in charts, and so do concepts and mechanisms such as how a score is built: a chart, a pipeline or a worked example beats rows of labels and sentences. Each headline number sits beside the figure that produced it. A source's official figure is used as is.
- Every mark carries data. Leave out decorative icons, big numerals with no quantity behind them, invented examples, side facts such as cost or release counts, one model's numbers shown alone, and categories with fuzzy or unclear edges.
- Signed values are drawn as magnitudes with the two directions labelled at the ends; the source's sign convention goes in the notes.
- A paper takes three slides: its first page with the question, one method figure, one comparison figure.
- One slide per method, index or example. The first slide of a new topic says what it is and how it works, with one real example.
- An index slide shows its composition and principles; strengths and weaknesses go in the notes.
- A slide shows everything at once, so the speaker clicks once per slide. A click build is kept for a reveal that is the point, such as the answer to a question the audience first guesses. A sequence is drawn as one still picture, such as the start faint under the end. Animate only when the user asks: the animation is then one continuous scene in an
<x-embed>, a canvas drawn as a function of time, each phase as long as the note lines said over it. - A leaderboard shows the top five with the uncertainty the source publishes, and its title names the interval.
- The cover states title, author and date. The agenda lists the parts, and the summary gives one point per part and a closing line.
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago Changed 8cac516eeec9
- 3d ago Changed · +1 lines ab7a08356449
- 23d ago Changed · +4 lines c5e86d598335
- 24d ago First seen · 46 lines · 119 tokens per session scan A d4fc600b0c6c
talk-slides is a skill published in the GitHub repository wenhaochai/claude-plugins (17 stars, last pushed 2d ago), licensed MIT. It adds 119 tokens to every session and 1,567 once invoked, about $0.0005 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-24.
Other skills, from other repositories
orbit-notion
Open Orbit briefing skill — selected by the Orbit pipeline when Notion is the user's only connected connector, or when the user explicitly scopes their daily digest to Notion. Pulls the past 24 hours of document edits, comments, mentions, and database row changes from the user's authenticated Notion connection and…
instrument-data-to-allotrope
Convert laboratory instrument output files (PDF, CSV, Excel, TXT) to Allotrope Simple Model (ASM) JSON format or flattened 2D CSV. Use this skill when scientists need to standardize instrument data for LIMS systems, data lakes, or downstream analysis. Supports auto-detection of instrument types. Outputs include full…
legal-hold
Issue, refresh, release, or report on legal holds — drafts the hold notice as .docx, updates legalhold fields in log.yaml, and calendars the next refresh. Use when the user says "issue a hold", "refresh hold", "release hold", or asks for a portfolio-wide hold status report.
baoyu-youtube-transcript
A tool for downloading the written captions, subtitles, chapter information, speaker labels, and cover image from a YouTube video using its URL or ID.
feishu
Work with Feishu or Lark bots, docs, sheets, bitables, approval flows, and OpenAPI/MCP setup without hardcoding credentials.
read
Fetches URLs and PDFs, then summarizes or returns clean Markdown. Use when asked to read, fetch, quote, cite, convert, or save a URL or PDF. Not for local text files already in the repo.