v1.2.0 — Sticky Toast + Keep-Alive

Fixed

• plugins/llm-statusline.toast.ts: toast is now sticky (duration=0 = never auto-dismisses). A keep-alive setInterval re-fires the toast every 4 min as a safety net, so the bar never disappears — even between model responses.
• Both plugins: version field was hardcoded to "opencode" (rendering 📟 vopencode). Now spawns opencode --version at init time.

Added

• LLM_STATUSLINE_TOAST_MS env var to override toast duration (default 0 = sticky).
• client.app.log calls for structured logging.
• Keep-alive timer (setInterval every 4 min).

Changed

• TOAST_MS defaults to 0 (sticky, never auto-dismiss).

v1.1.0 — Toast Variant + Vendored Python

Added

• New plugins/llm-statusline.toast.ts server plugin that emits the rendered bar as a TUI toast on every session.idle, instead of writing to the log panel.
• Vendored python/ directory: a self-contained copy of the upstream statusline script. The plugin no longer depends on any Claude Code install at the filesystem level.
• OPENCODE_PROJECT_DIR environment variable as an optional override for the folder displayed in the toast.

Changed

• Cache directory moved from ~/.cache/llm-quota-bar/ to ~/.cache/opencode-llm-statusline/ to keep state separate from upstream.
• .gitignore now ignores python/__pycache__/.

Known Limitations

• OpenCode 1.17.8 plugin SDK does not expose tool-call parts; the toast reflects the shell cwd, not the live project the agent is editing.
• OpenCode's tui.showToast does not render ANSI escape codes; colors are stripped before display.

v1.0.0 — Initial Release

Added

• Initial release (plugins/llm-statusline.ts).
• Forwards OpenCode session.idle events to a session_tokens.py script, writing a Claude-Code-compatible JSONL so the upstream statusline logic (token totals, cache R/W, burn rate, cost, provider quota) can be reused unchanged.
• Logs the rendered bar through client.app.log() (visible in OpenCode's log panel).