repospend 0.0.7 → 0.0.9

package/CHANGELOG.md

@@ -2,9 +2,33 @@
 
 All notable changes to RepoSpend will be documented in this file.
 
+## 0.0.9
+
+- Cache parsed Codex session summaries under `~/.repospend/cache` so unchanged large transcripts reload much faster.
+- Limit local source scanning to the active dashboard date window when possible, reducing default last-7-days scan work.
+- Warm a bounded last-30-days cache in the background after the default last-7-days dashboard loads, without auto-scanning all-time history.
+- Make refresh/rescan/retry actions clear RepoSpend's parse cache before reloading local logs.
+- Add an Advanced settings action to clear only the parse cache and reload without deleting pricing or source settings.
+- Keep demo mode from reading, warming, or clearing the real local parse cache.
+- Make experimental Cursor support opt-in so default scans and dashboard surfaces stay focused on Codex and Claude Code unless Cursor is explicitly enabled in Settings.
+- Add a local Data Sources setting for enabling Cursor, with copy explaining that reliable Cursor tokens, model names, and costs are often unavailable from local transcript files alone.
+- Hide Cursor quick links, loading-state rows, and status copy while the experimental Cursor source is disabled.
+- Improve Cursor CLI metadata recovery by inferring transcript dates and repo paths from Cursor's local project/transcript paths when records omit timestamp or cwd fields.
+- Avoid an unnecessary standalone Codex session pass for sessions already imported from the Codex SQLite thread index.
+
+## 0.0.8
+
+- Make `repospend serve` and `npx repospend` recover when the default localhost port is already in use by trying nearby ports instead of exiting with `EADDRINUSE`.
+- Print a clear message when RepoSpend falls back from the requested port to the actual dashboard port.
+- Add a Models view with per-model token shape, API-equivalent cost, cache reuse, top repo, latest activity, and one-click filter into Sessions.
+- Reorder the Overview so the Top repositories and Waste signals panels appear higher on the page.
+- Refresh the dashboard accent color from purple to teal across nav, KPI cards, and the metric timeline.
+- Replace the Sessions page mini-stats with filter-aware totals (tokens, API-equivalent cost, sessions needing review, edits, commands).
+- Make the filters bar scroll with the page instead of sticking to the top.
+- Collapse the "How Agent Friction is classified" explainer into a disclosure to reduce vertical noise on the Agent Friction page.
+
 ## 0.0.7
 
-- Bump RepoSpend package versions to 0.0.7.
 - Add experimental Cursor support with read-only discovery for local JSONL transcripts and SQLite/vscdb storage, keeping sessions visible when token or cost data is unavailable.
 - Include Cursor in source filters, dashboard source health, README data-source documentation, and CLI help.
 - Track known vs. unknown-cost sessions in grouped breakdowns so mixed Cursor or Claude data does not hide partial cost coverage.

package/README.md

@@ -68,11 +68,12 @@ prints the dashboard URL. By default it runs at
 |---|---|---|
 | Codex | Most complete support | Tokens, models, sessions, repo grouping, command friction |
 | Claude Code | Initial support | Sessions, projects, models, timestamps, tokens when available |
-| Cursor | Experimental | Local JSONL and SQLite/vscdb discovery; tokens/cost only when local data includes them |
+| Cursor | Experimental opt-in | Local JSONL and SQLite/vscdb discovery; tokens/cost only when local data includes them |
 | RTK | Optional/local | Shown only when local RTK data exists |
 
-RepoSpend started as a Codex-first release. Claude Code and Cursor support are
-newer and depend on what those tools persist locally.
+RepoSpend started as a Codex-first release. Claude Code support is newer, and
+Cursor is off by default because accurate Cursor token usage often requires
+account-backed usage data rather than local files alone.
 
 ## Requirements
 
@@ -121,6 +122,13 @@ token intensity across projects.
 Repo detail explains why a project stands out, including cost concentration,
 warnings, token shape, sessions, and command signals.
 
+### Models
+
+![RepoSpend models view with fictional model usage and token shape](docs/screenshots/models-view.png)
+
+The models view compares token shape, API-equivalent cost, cache reuse, sessions,
+and repo concentration across the models used in the current scan.
+
 ### Sessions
 
 ![RepoSpend sessions table with fictional session titles](docs/screenshots/sessions-view.png)
@@ -239,12 +247,18 @@ RepoSpend-owned settings:
 ```text
 ~/.repospend/pricing.json
 ~/.repospend/config.json
+~/.repospend/cache/
 ```
 
 The Settings page includes a reset action for RepoSpend-owned files under
 `~/.repospend/`. It does not delete or edit anything under `~/.codex` or
 `~/.claude`.
 
+RepoSpend caches parsed session summaries under `~/.repospend/cache/` so
+unchanged large transcripts reload faster. Rescan actions clear this parse cache
+before reading local logs again. The Advanced settings tab can clear only the
+parse cache without removing pricing or source settings.
+
 ## Commands
 
 Most users only need:
@@ -286,8 +300,9 @@ browser.
   token usage are shown when present in local JSONL files.
 - Claude Code sessions without local token details are shown with unknown
   tokens/cost.
-- Cursor support is experimental: local transcript/session discovery is
-  best-effort, and Cursor may omit token/cost details or change local schemas.
+- Cursor support is experimental and off by default: local transcript/session
+  discovery is best-effort, and Cursor may omit token/cost details or change
+  local schemas.
 - Cost estimates do not represent subscription billing, credits, regional
   pricing, or account-specific terms.
 - Budget alerts are not available yet.