Awesome-Agent-Skills-for-Empirical-Research talk
Create and audit presentations (Beamer or Quarto RevealJS). Combines talk creation, visual audit, and compilation. Replaces /create-talk, /visual-audit, /compile-latex (for talks).
git clone https://github.com/brycewang-stanford/Awesome-Agent-Skills-for-Empirical-Research
T=$(mktemp -d) && git clone --depth=1 https://github.com/brycewang-stanford/Awesome-Agent-Skills-for-Empirical-Research "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/16-hsantanna88-clo-author/dot-claude/skills/talk" ~/.claude/skills/brycewang-stanford-awesome-agent-skills-for-empirical-research-talk && rm -rf "$T"
skills/16-hsantanna88-clo-author/dot-claude/skills/talk/SKILL.mdTalk
Create, audit, or compile presentations (Beamer or Quarto RevealJS).
Input:
$ARGUMENTS — mode and format/path.
Modes
/talk create [format]
— Create Beamer Talk
/talk create [format]/talk create [format] --quarto
— Create Quarto RevealJS Talk
/talk create [format] --quartoGenerate a presentation from the paper.
Agents: Storyteller (creator) → storyteller-critic (reviewer)
Format Constraints
| Format | Slides | Duration | Content Scope |
|---|---|---|---|
| job-market | 40-50 | 45-60 min | Full story, all results, mechanism, robustness |
| seminar | 25-35 | 30-45 min | Motivation, main result, 2 robustness, conclusion |
| short | 10-15 | 15 min | Question, method, key result, implication |
| lightning | 3-5 | 5 min | Hook, one result, so-what |
Workflow
Step 1: Parse Arguments
- Format (required):
|job-market
|seminar
|shortlightning - Paper path (optional): defaults to
paper/main.tex - Engine: Beamer (default) or Quarto RevealJS (
)--quarto - If no format specified, ask the user.
Step 2: Dispatch Storyteller
Read the paper and extract: research question, identification strategy, main result, secondary results, robustness checks, key figures/tables, institutional background. Design narrative arc for the chosen format. Build the slide file with shared preamble if available.
The Storyteller follows these design principles:
- One idea per slide — never cram two concepts onto one frame
- Figures over tables; tables in backup — audiences absorb figures instantly; regression tables belong in backup slides where referees can inspect them during Q&A
- Build tension — motivation → question → method → findings → implications
- Transition slides between major sections — signal where the talk is going
- All claims must appear in the paper — the paper is the single source of truth; never add results or claims that are not in the manuscript
Compile with XeLaTeX (Beamer) or
quarto render (Quarto).
Save to
paper/talks/[format]_talk.tex (Beamer) or paper/quarto/[format]_talk.qmd (Quarto).
Step 3: Dispatch Storyteller-Critic
After the Storyteller returns, dispatch the storyteller-critic to review across 5 categories:
| Category | What It Checks |
|---|---|
| Narrative flow | Does the story build properly? Is there a clear arc from motivation through results to implications? Are transitions smooth? |
| Visual quality | Text overflow, font readability (>= 10pt), figure sizing, consistent formatting, overfull hbox warnings |
| Content fidelity | Every claim traceable to the paper — no orphan results, no unsupported statements |
| Scope for format | Right amount of content for the duration — not cramming a seminar into a lightning talk, not padding a short talk to seminar length |
| Compilation | Does it compile cleanly without errors or warnings? |
Score as advisory (non-blocking). Save report to
quality_reports/[format]_talk_review.md.
Step 4: Fix Critical Issues
If the storyteller-critic finds Critical issues (compilation failures, content not in paper):
- Re-dispatch Storyteller with specific fixes (max 3 rounds per three-strikes rule)
- Re-run storyteller-critic to verify
Step 5: Present Results
Report to the user:
- Generated file path
- Slide count and format compliance
- Storyteller-critic score (advisory, non-blocking)
- TODO items (missing figures, tables not yet generated)
/talk audit [file]
— Visual Audit
/talk audit [file]Check existing slides for layout issues.
Run visual quality checks:
- Text overflow on any slide
- Font sizes (>= 10pt for projection)
- Table readability
- Figure sizing and labels
- Consistent formatting
- Overfull hbox warnings
/talk compile [file]
— Compile Talk
/talk compile [file]3-pass XeLaTeX compilation for Beamer:
cd paper/talks && TEXINPUTS=../preambles:$TEXINPUTS xelatex -interaction=nonstopmode [file]
For Quarto:
cd paper/quarto && quarto render [file]
Principles
- Paper is authoritative. Every claim must appear in the paper.
- Figures over tables. Audiences absorb figures instantly. Put regression tables in backup slides for Q&A.
- Less is more. Especially for short and lightning formats — ruthlessly cut.
- One idea per slide. If you need a second point, make a second slide.
- Audience calibration. Job market = demonstrate rigor and command of the literature. Seminar = sell the interesting result. Short = method and key finding. Lightning = sell the idea in one breath.
- Advisory scoring. Talk scores don't block commits.
- Worker-critic pairing. Storyteller creates, storyteller-critic critiques. Never skip the review.