CtrlK
BlogDocsLog inGet started
Tessl Logo

jbaruch/speaker-toolkit

Four-skill presentation system: ingest talks into a rhetoric vault, run interactive clarification, generate a speaker profile, then create new presentations that match your documented patterns. Includes an 88-entry Presentation Patterns taxonomy for scoring, brainstorming, and go-live preparation.

96

1.21x
Quality

93%

Does it follow best practices?

Impact

97%

1.21x

Average score across 30 eval scenarios

SecuritybySnyk

Advisory

Suggest reviewing before use

Overview
Quality
Evals
Security
Files

task.mdevals/scenario-24/

Plan a YouTube Thumbnail for a Delivered Talk

Problem/Feature Description

A speaker delivered "Robocoders: Judgment Day" at DevNexus 2026 two weeks ago. The recording is now on YouTube at https://youtube.com/watch?v=abc123. The speaker wants a YouTube thumbnail that will maximize click-through rate.

The outline has an illustration strategy (retro sci-fi propaganda poster aesthetic) with several full-bleed illustration slides, plus text slides, a bio slide, a shownotes slide, and bullet-heavy slides. The speaker's profile has thumbnail preferences configured.

Using the presentation-creator skill, plan the complete thumbnail generation workflow: select candidate slides, propose a title, and document the exact command to generate the thumbnail.

Output Specification

Produce the following files:

  1. thumbnail-plan.md — Complete thumbnail workflow plan including:
    • List of 3-5 candidate slides ranked by visual impact, with reasoning for each
    • Which slides were excluded and why
    • 2-3 proposed title text options (short hooks, not the full talk title)
    • The exact generate-thumbnail.py command that would be run (with all arguments)
    • Where the speaker photo comes from
    • How face preservation is handled in the generation process
  2. slide-candidates.md — Formatted list of candidate slides with:
    • Slide number and description
    • Why this slide is a good/bad candidate
    • Source for the slide image (illustration file, extraction, etc.)

Input Files

Download vault fixtures from the project repository:

BASE="https://github.com/jbaruch/speaker-toolkit/raw/main/eval-resources/scenario-24"
mkdir -p inputs/vault inputs/talk
curl -L -o inputs/vault/speaker-profile.json "$BASE/speaker-profile.json"
curl -L -o inputs/vault/secrets.json "$BASE/secrets.json"
curl -L -o inputs/vault/tracking-database.json "$BASE/tracking-database.json"
curl -L -o inputs/talk/presentation-spec.md "$BASE/presentation-spec.md"
curl -L -o inputs/talk/presentation-outline.md "$BASE/presentation-outline.md"

Key Parameters

  • YouTube URL: https://youtube.com/watch?v=abc123
  • Talk title: "Robocoders: Judgment Day"
  • Illustration style: Retro sci-fi propaganda poster
  • Illustration slides: 1, 10, 25 (full-bleed illustrations)
  • Bio slide: 4 (should be avoided)
  • Shownotes slide: 3 (should be avoided)
  • Bullet-heavy slide: 18 (should be avoided)
  • Profile thumbnail preferences: style=overlay, title_position=bottom, brand_colors=#5B2C6F,#C0392B
  • Speaker photo path: configured in profile

evals

README.md

tile.json