v0.12.19

Share Export\n- Rebuild conversation share export with HTML + WebKit rendering\n- Fix truncated content and PNG/PDF pagination/cropping issues\n- Add clearly distinct visual themes for Balanced/X/LinkedIn/Slack/Telegram\n\n## macOS App\n- Restore status bar usage text to white for better readability\n- Adjust panel/floating behavior to avoid interfering with Dock auto-show\n\n## Build\n- Link WebKit framework in app compile scripts for local/dev and wheel builds

v0.12.18

Pricing\n- Recalibrate model pricing with latest official rates for OpenAI, Claude, and Gemini\n- Add dedicated pricing matcher for Codex/OpenAI model families (including gpt-5.3-codex)\n- Unify Python-side cost estimation across reporter/webhook/hooks to avoid drift\n\n## Accuracy\n- Improve Claude cache savings estimation with per-model price delta\n- Keep Swift app pricing map aligned with CLI/backend behavior

v0.12.17

Performance\n- Reduce macOS app memory pressure in parser/analyzer path\n- Add progressive dashboard loading (fast first batch + background completion)\n- Add huge-dataset guardrail (prioritize recent 30-day sessions, then backfill history on demand)\n\n## UX\n- Conversation panel now lazy-loads full session messages when opened\n\n## Quality\n- Test suite passed: 284/284

v0.12.16

Bug Fixes

• Add Codex sessions support for ~/.codex/sessions/YYYY/MM/DD/rollout-*.jsonl in CLI/reporter/export/webhook.
• Fix Codex token attribution and deduplicate cumulative token_count events.
• Fix Codex model attribution from turn_context.payload.model so model token breakdown no longer shows unknown.
• Fix macOS app startup in editable/dev mode to recompile Swift sources when they change.
• Fix macOS app Codex parser field mapping so Codex token usage is displayed correctly.

Version

• Bump version to 0.12.16 and sync version files.

v0.12.14

What's New

New Features

• Rate Limit Predictor — Safe/Warning/Critical status with headroom estimate (--rate-limit)
• Cache Efficiency Grading — Excellent/Good/Fair/Poor with savings estimate
• Git Integration — AI cost attribution per commit (--git)
• Coding Rhythm Analysis — Morning/Afternoon/Evening/Night usage patterns
• Work Mode Classification — Exploration/Building/Execution auto-detection

Bug Fixes

• Fixed negative usage time display (both CLI and macOS App)
• Fixed table alignment for CJK characters in --compare output

UX Improvements

• macOS App: 'API Token' renamed to 'Session Token' with clearer instructions
• macOS App: Token expiry detection with visual indicator
• macOS App: Rate limit fetch throttled to 60s (faster loading)
• macOS App: Skill stats cached (reduced redundant computation)

Internal

• 95 new unit tests (262 total)
• README_EN.md rewritten with comparison table and screenshots

v0.12.13

Bug Fixes

• fix: subagent sessions double-counted — exclude subagents/ directory sessions to prevent token double-counting (#20)
• fix: --since/--until date filter precision — corrected date range boundary calculation (#20)

v0.12.12

New Features

• Auto-upgrade — When a new PyPI version is detected, automatically runs pipx upgrade cc-statistics in the background (fallback to pip). Sends system notifications for upgrade progress

Recent Fixes (v0.12.11)

• Include subagent token usage — Fixed ~50% token underreporting by scanning subagent JSONL files

v0.12.11

Bug Fixes

• Include subagent token usage in statistics — SessionParser now scans <session-id>/subagents/*.jsonl files and merges their messages into the parent session. Previously subagent API calls were uncounted, causing ~50% token underreporting

v0.12.10

Bug Fixes

• Fix loading state stuck forever (#19) — Use defer for isLoading reset in all Task entry points (init/refresh/setTimeFilter), ensuring loading state is always cleared even if data parsing fails or exits unexpectedly

v0.12.9

New Features

• Usage Burn Rate Monitor — 3-level alert system (info/warning/critical) that tracks consumption speed across 5h and 7d windows. Uses sliding window smoothing and debounce to prevent false alarms. Dashboard gauge color changes by alert level (#19)

Bug Fixes

• Model pricing update — Opus 4.5/4.6 standard corrected to $5/$25 (was $15/$75), Haiku 4.5 added at $1/$5. Aligned with official pricing table

Documentation

• Cross-platform data capability matrix — README now shows what data is available from each platform (Claude Code / Codex / Gemini / Cursor)