A Claude Code skill that takes today's top headlines, runs Karpathy's iterative argument/counter-argument loop on each one, and delivers a balanced briefing so you can read both sides and form your own conclusions.
Inspired by this exchange between Chamath Palihapitiya and Andrej Karpathy.
Copy the skill file into your Claude Code project:
mkdir -p .claude/commands
curl -o .claude/commands/dialectic-digest.md \
https://raw.githubusercontent.com/Percival-Labs/dialectic-digest/main/dialectic-digest.mdOr just download dialectic-digest.md and drop it in .claude/commands/.
/dialectic-digest
Get today's top 5-7 headlines with full dialectic analysis.
/dialectic-digest AI regulation
Focus on a specific topic.
For each headline, the skill runs two full rounds of argument and rebuttal:
- Thesis — The strongest case for the mainstream position
- Antithesis — An equally rigorous counter-argument
- Thesis Rebuttal — Responds to the counter-argument, concedes where warranted
- Antithesis Rebuttal — Responds back, concedes where warranted
Then a synthesis that identifies:
- Where both sides actually agree
- The core factual disagreement (what data would settle it?)
- The core values disagreement (what priorities are in tension?)
- What to watch for next
The iteration is the key insight. A single pass gives you "here's side A, here's side B." Two rounds of rebuttal force each side to engage with the other's strongest points and concede where they're wrong. What survives is worth reading.
See example-output.md for a full run on the Meta/YouTube social media addiction verdict (March 28, 2026).
- Claude Code (any version with web search)
- No API keys, dependencies, or setup beyond Claude Code itself
It's one markdown file. The file contains structured instructions that Claude follows when you invoke the command. Claude uses its built-in web search to find today's headlines, then runs the dialectic loop using its own reasoning. No external services, no backend, no database.
MIT. Do whatever you want with it.
Built by Percival Labs. We build trust infrastructure for the agent economy.