Creates new Claude Code agent configuration files or audits existing ones in .claude/agents/. Validates frontmatter, routing keywords, body-frontmatter alignment, ecosystem integration, and cross-agent consistency using a 17-point checklist.
Look-ahead bias prevention and portfolio formation rules for cross-sectional asset pricing factors. Covers signal timing, portfolio sorts, return alignment, and rebalancing conventions. Auto-apply when constructing factors, sorting stocks into portfolios, or computing long-short returns.
Adversarial research idea generator for empirical asset pricing. Surveys literature via Perplexity, stress-tests hypotheses, maps WRDS data feasibility, and compiles a research plan skeleton. Use when brainstorming new paper ideas, sharpening existing hypotheses, or resuming a prior ideation session.
Create a new empirical research project: generates projects/ / with latex, code, scripts, results, literature, and guidance subfolders plus README.md and project-level CLAUDE.md.
Agent-driven cold-start onboarding. Use the repo shell entrypoint to find or install Python 3.11+, then run the shared bootstrap audit/plan/apply flow and configure WRDS only if the user has it.
Rules and gotchas for CRSP, Compustat, and financial panel data: safe lagging/leading with date gap checks, look-ahead bias prevention, CCM linking, book equity construction, and missing value conventions. Auto-apply when working with CRSP, Compustat, or financial panel data.
Publication-ready figure conventions for empirical finance and economics. Covers matplotlib styling, color palettes, export settings, and common figure types (time series, decile bars, coefficient plots, event studies). Auto-apply when creating any figure, plot, or visualization.
Generates structured results reports after PyBondLab portfolio formation runs. Auto-apply when running StrategyFormation, BatchStrategyFormation, BatchWithinFirmSortFormation, or any portfolio sorting with PyBondLab.
Fast batch/single portfolio sort on Dickerson bond data. Bypasses orchestrator agent — runs directly via Bash. Use for known workflows on real data. Examples: /run batch cs ytm bbtm, /run batch cs ytm --wfs, /run single cs --nport 10.
Scaffolds a new academic paper from the LaTeX boilerplate template. Copies template files, replaces placeholder tokens with user-provided title, authors, and affiliations, writes [REMOVE]-tagged filler content with Fama-French citations, and compiles a test PDF. Use when starting a new paper or creating a fresh LaTeX…
Download, split, and deeply read academic PDFs with adaptive chunking. Splits into optimally-sized chunks based on paper length, reads in controlled batches, and produces structured extraction notes tailored for empirical finance papers.
Detect documentation drift and propose updates to docs/ai/, AGENTS.md, and CLAUDE.md. Use when session-start warnings flag stale docs, after significant code changes, or periodically to keep context current.