Setup LLM Optimization

What This Sets Up

• SessionStart hook setting AI_AGENT=1, CLAUDECODE=1, and NODE_OPTIONS=--max-old-space-size=8192
• PreToolUse hook optimizing test commands:
  • Strip --verbose (hard enforcement via updatedInput rewrite — wastes tokens)
  • Suggest --pool=forks (prevents zombie processes)
  • Suggest --bail=1 (fail fast, save tokens)
  • Suggest --teardownTimeout=5000 (prevent hanging teardown)
  • Suggest --reporter=github in CI
  • Jest: suggest --bail --forceExit
• PostToolUse hook truncating verbose bash output to reduce context bloat

Steps

1. Create hook scripts

Copy scripts/llm-env.sh, scripts/llm-test-flags.sh, and scripts/llm-truncate.sh into .claude/hooks/. Make all executable.

2. Configure hooks in .claude/settings.json

Add to hooks config (merge with existing):

• SessionStart: .claude/hooks/llm-env.sh
• PreToolUse (matcher: Bash): .claude/hooks/llm-test-flags.sh
• PostToolUse (matcher: Bash): .claude/hooks/llm-truncate.sh

3. Verify

• All hook scripts are executable
• AI_AGENT and CLAUDECODE are set after session start
• bun test --verbose is silently rewritten to bun test (verbose stripped)
• bun test triggers suggestions for --pool=forks, --bail=1, --teardownTimeout=5000
• NODE_OPTIONS=--max-old-space-size=8192 is set after session start
• Long output is truncated

See REFERENCE.md for vitest config optimizations and test runner tuning.

4. Commit

Stage and commit: Add LLM optimization hooks (env vars, test flags, output truncation)