v1.1.6

Fixed

• ai-resources audit rewritten from scratch — the previous implementation
  called litellm.container_logs() (a function that never existed) and expected
  LiteLLM to emit cost lines in a specific log format it never produced. The
  command crashed immediately on every invocation.

  The new implementation reads directly from Claude Code's JSONL session files
  (~/.claude/projects/<slug>/*.jsonl), which contain the authoritative token
  usage and model name for every API call made by the cockpit and all subagents.

  New flags:

  • --days N — limit report to last N days (default: all time)
  • --verbose / -v — show per-session breakdown table (date, model, calls,
    output, cache-read, cost)

  Cost estimates use a built-in price table (Anthropic + Google public list
  prices). Gemini calls routed via LiteLLM appear with their correct provider
  model name when the gateway returns it, confirming delegation is working.