Autonomous quality iteration loop. Modifies pipeline code, runs experiments, keeps improvements, reverts regressions. Loops until interrupted, bounded with /loop N, or until a finish condition is met. Supports single-scorer targeting or sweep mode across all scorers. Inspired by Karpathy's autoresearch.
Run the full Astral Index weekly publishing pipeline with quality gates. Scrapes sources, generates drafts, evaluates quality, and delivers via Buttondown. Use when publishing a newsletter issue.
Quality improvement dev loop. Identifies weak scorers, proposes a change, implements it, runs a Braintrust experiment, and compares scores. Use when improving newsletter quality or trying a specific pipeline change.
Audit the evaluation system. Generates a test draft, independently assesses quality, compares to scorer output, and identifies blind spots or miscalibrations. Use to validate scorers are measuring what matters.
Strategic brainstorming for newsletter quality. Analyzes the pipeline holistically, identifies gaps in source coverage, and proposes new features, strategies, and scorer dimensions. Use for high-level quality improvement ideas.