Production engineering.
Now agentic.

# Read the files, don't just skim the index.

# Why
#   Every new AI session auto-loads the index, but NOT
#   the rule files linked from it. The session sees a
#   one-line summary and thinks it knows the rule.
#   Summaries are pointers; the rule lives in the file.

# How the setup enforces it
#   1. The index (docs/context/INDEX.md) auto-loads
#      on every new session via the project's CLAUDE.md
#      import.
#   2. Each entry links to a rule file with a summary
#      that ends "read the file, not just the line."
#   3. Before any diff, copy change, or factual claim,
#      the session reads every rule whose summary
#      overlaps the user's current topic.
#   4. Writes only start after that reconciliation pass
#      completes. No exception for "obvious" requests.

# What goes wrong without it
#   Session skims an index entry, generates code or
#   copy that silently violates the rule, ships. The
#   failure surfaces in review (best case) or in
#   production (worst case). Cost compounds because the
#   next session re-makes the same skim.

# Enforcement
#   - CLAUDE.md flags this rule by name on turn one
#   - INDEX.md summaries explicitly say "read the file"
#   - Every observed violation produces a new rule;
#     the trap codifies itself