CLAUDE.md

CLAUDE.md — Project Conventions

File Operations

• When asked to edit a file, use Glob/Grep to locate it rather than asking the user for the path.
• Prefer targeted reads (offset + limit) over reading entire large files.
• Before creating a new file, check whether a suitable one already exists.
• Do not re-read a file after editing it to verify — edits would have errored if they failed.

Editing Conventions

• When editing R/Quarto files, check for stale object/column name references after renames (e.g., title -> new name) across the entire file, not just the edit site.
• Make the minimal change needed — do not refactor or clean up surrounding code unless asked.
• Do not add comments, docstrings, or type annotations to code you did not change.
• Avoid backwards-compatibility shims for removed code; delete unused code outright.

R Conventions

• This project uses data.table, not data.frame/dplyr. Always use data.table syntax (e.g., DT[, .(col)], :=) and verify column access patterns work on data.table objects before finalizing edits.
• Use set* functions (setnames, setorder, setkey) for in-place mutations to avoid copies.
• Prefer fread/fwrite over read.csv/write.csv.
• Do not load tidyverse or dplyr unless explicitly requested.

Performance / Memory

• Large datasets hit R memory limits — prefer chunking, avoid unnecessary copies, and add defensive guards for list vs data.table structures before rbindlist.
• Never use rbind in a loop; accumulate results in a list and call rbindlist once.
• Avoid apply-family functions on large data.tables; use vectorized data.table operations instead.
• Check object.size() / lobstr::obj_size() when debugging memory issues rather than guessing.

Token Efficiency

• Do not summarize what you just did at the end of a response — the diff is visible.
• Do not repeat back the user's request before answering.
• Skip pleasantries and filler phrases ("Great question!", "Certainly!", etc.).
• When reading code to answer a question, read only the relevant section, not the whole file.
• Use files_with_matches output mode for Grep unless line content is needed.