# Lockfiles are managed by their package manager — never reformat them.
pnpm-lock.yaml
package-lock.json
yarn.lock

# Build output and generated artifacts
# Root CHANGELOG.md carries sections generated byte-for-byte by
# scripts/release/generate-release-notes.mjs — reformatting breaks its idempotent regeneration.
/CHANGELOG.md

# Auto-generated lesson artifacts, on the same grounds as CHANGELOG.md above. `lessons-lib.mjs`
# rewrites both on every push, `.husky/pre-commit` refuses to commit them, and the checked-in
# `weekly-digest.md` is already prettier-unclean — so `format-check`, the one local stage no required
# CI check re-runs, failed on two files the person running it is not allowed to fix. A guard firing
# on correct work is how a guard stops being read (INFRA-066).
.agents/evals/lessons/auto-lessons.md
.agents/evals/lessons/weekly-digest.md
dist/
**/dist/
coverage/
.next/

# D2: common-mistakes.md is a 88-row catalogue. Prettier's table alignment padded it to 195 KB, of
# which 156 KB (80%) was alignment whitespace — and AGENTS.md tells every agent to read this file
# before non-trivial work, so that padding was paid as context on every read. Squeezed to 37 KB.
# Safe because scan-mistake-mechanisms parses the `**Mechanism:**` TOKEN and a `| <n> |` row opener
# with `\s*`, never column geometry: verified by re-running it across the squeeze — 88 entries,
# 39 mechanisms, 49 accepted, before and after.
.agents/rules/common-mistakes.md
