CtrlK
BlogDocsLog inGet started
Tessl Logo

jbaruch/speaker-toolkit

Five-skill presentation system: ingest talks into a rhetoric vault, run interactive clarification, generate a speaker profile, create presentations that match your documented patterns, and produce the deck illustrations + thumbnail visual layer. Includes a 102-entry Presentation Patterns taxonomy (91 observable, 11 unobservable go-live items) for scoring, brainstorming, and go-live preparation.

95

1.20x
Quality

96%

Does it follow best practices?

Impact

95%

1.20x

Average score across 32 eval scenarios

SecuritybySnyk

Advisory

Suggest reviewing before use

Overview
Quality
Evals
Security
Files

task.mdevals/scenario-32/

Pick the Right Thumbnail Aesthetic from Vault Signals

Problem/Feature Description

A speaker delivered "Never Trust a Monkey" at JCON Europe 2026 last week. The recording is on YouTube. The speaker asks for a thumbnail and provides the YouTube URL — that's the entire request. They have no opinion on aesthetic; they want the agent to recommend one.

Pick the recommended aesthetic from the speaker's profile, generate the thumbnail, and document the reasoning so a reviewer can audit the pick.

Output Specification

Produce two files:

  1. aesthetic-decision.md — The decision write-up. Must include:
    • Which profile field produced the recommendation, with the exact value quoted from the profile.
    • Which earlier-precedence fields were checked first and not used (so the audit trail is complete).
    • The recommended --aesthetic value.
    • A one-line note on whether a side-by-side two-candidate comparison is offered.
  2. generate-command.sh — The generate-thumbnail.py invocation that the agent would run, with all arguments substituted from the profile.

Input Files

The following file is provided. Extract it before beginning.

=============== FILE: inputs/speaker-profile.json =============== { "schema_version": 1, "generated_date": "2026-04-15", "talks_analyzed": 18, "speaker": { "name": "Dr. Maya Patel", "handle": "@mpresearch", "website": "https://mpresearch.dev", "shownotes_url_pattern": "https://mpresearch.dev/{slug}" }, "publishing_process": { "thumbnail": { "speaker_photo_path": "/Users/maya/photos/headshot-2026.jpg", "style_preference": "slide_dominant", "title_position": "top", "brand_colors": ["#1F3A5F", "#E8B04B"] }, "shownotes": { "enabled": false } }, "visual_style_history": { "default_illustration_style": "comic-book halftone with bold ink outlines", "style_departures": [ { "talk_slug": "ml-fairness-2024", "style": "watercolor diagram", "trigger": "academic audience" } ], "mode_visual_profiles": [ { "mode_id": "viral_keynote", "default_aesthetic": "comic_book" } ], "confirmed_visual_intents": [ { "pattern": "halftone shading on every slide", "rule": "comic-book caricature for any speaker portrait" } ] } } =============== END FILE ===============

The talk's working directory contains:

  • presentation-outline.md (no Illustration Style Anchor section for this particular talk).
  • illustrations/slide-08.png — the illustrated slide the speaker chose for the thumbnail (already provided).
  • The YouTube URL: https://youtube.com/watch?v=monkey2026.
  • The hook title: the speaker has confirmed "NEVER TRUST A MONKEY".

evals

README.md

tile.json