v0.1.8

Chat UX & bug fixes

• Message timestamps (sv-SE 24h) and hover tooltips for tool calls (<abbr>)
• Cache-token tracking and compression notice only on ≥30% reduction
• Approval flow: state queue, batch-resolve all pending, survive reload, don't auto-deny on new messages
• Fix dangling tool_use when parallel destructive calls needed approval
• Fix stuck "Running..." on parallel tool calls
• Scope Tailwind to .gds-assistant via postcss-prefix-selector (no admin bleed)
• WP-CLI: wp gds-assistant audit list|conversations|show|export
• Bulk-operation guidance added to system prompt

LLM providers

• Opus 4.7 (claude-opus-4-7)
• Anthropic prompt caching on conversation history
• Gemini functionResponse always a dict (fixes "cannot start list" errors)
• OpenAI: strict schema fix (properties default for type: object) and max_completion_tokens for o1/o3/o4/gpt-5 reasoning models
• Remove 20k-char JSON truncation (ContextCompressor handles oversized results)