feat: session stats, optional two-line wrap, mode glyphs — happy to split into PRs

[rondiver]

Using the tool daily — thanks for building it. I've prototyped four enhancements locally and would like to upstream them. Happy to split into separate PRs to keep reviews manageable.

1. Session stats from transcript. Parses $input.transcript_path with jq to compute running cost (per-model price table for Opus/Sonnet/Haiku incl. cache read/write tiers), user/assistant message counts, and session duration. Caches the parse by transcript mtime so it only re-runs when the transcript changes. Appends $7.32 | msg 3/25 | 12h34m before the clock.

2. Optional two-line wrap. Opt-in via CLAUDE_STATUSLINE_WRAP=true (or a settings.json key). Splits after effort — context state on line 1, budget/session on line 2. Avoids the awkward soft-wrap on narrow terminals.

3. Mode glyphs. Reads alwaysThinkingEnabled and fastMode from settings.json, renders 🧠 / ⚡ after the model name. Nothing shown when disabled.

4. Compact-context warning. When token usage ≥85%, prefix the token segment with a red ⚠ — visual nudge before Claude Code auto-compacts.

Licensing question: The repo doesn't currently have a LICENSE file, so strictly speaking it's all-rights-reserved. Would you be open to adding one (MIT / Apache-2.0 / whatever you prefer)? I'd rather not send PRs that can't legally be merged or reused.

[daniel3303]

Thanks Ron — and glad it's useful.

License: the project has been MIT since day one (see the License section in the README) — I just hadn't added the actual LICENSE file. Fixed that now in #47, so contributions are unambiguously fine.

Splitting into separate PRs is exactly right. A few notes per item so you can scope them before you push:

(1) Session stats from transcript. Happy to take this, but the model price table is the part I want to get right before merging. Cache-read/write tiers and new models change often, so I'd rather it lived in a single data block at the top of the script (one source of truth, easy to bump) than be sprinkled through the parsing logic. Two more asks:

• Make it opt-in (STATUSLINE_SESSION_STATS=true), off by default — not everyone wants cost in their statusline.
• Treat it as additive to the existing usage segment, not a replacement. If you've prototyped a format, paste a screenshot in the PR so we can sanity-check it against the current layout before you polish it.

(2) Two-line wrap. Yes please. One nit: rename the env var to `STATUSLINE_WRAP=true` to match the existing `STATUSLINE_CHECK_UPDATES` convention (drop the `CLAUDE_` prefix).

(3) Mode glyphs. Yes. Two requirements: (a) opt-out via `STATUSLINE_GLYPHS=false` for terminals without emoji rendering, and (b) parity in `statusline.ps1` — every change has to land in both scripts.

(4) Compact-context warning at ≥85%. Yes, simple and useful. 85% is a fine default; happy for it to be a constant rather than configurable unless it's trivial to expose.

Cross-cutting: keep both `statusline.sh` and `statusline.ps1` in sync in the same PR, and avoid adding runtime deps beyond `jq` (already used). Looking forward to the PRs.