v1.1.0 - Report Card

tokenuse 1.1.0, named Report Card, adds a Coach page to the desktop app: a report-card workspace that grades how you work with your AI tools, backed by 27 locally evaluated practice rules, a richer call archive, and deeper per-tool parsers. Everything is computed from the usage files already on disk — no network calls, no LLM in the loop.

Highlights

• A new Coach sidebar page (shortcut h) with report-style tabs for Report, Findings, AI Output, and Activity.
• Detection is a Rust-native port of 27 rules from AI-Engineering-Coach, Microsoft's MIT-licensed VS Code extension (see NOTICE and docs/development/coach.md), evaluated entirely locally.
• The call archive moves to schema v4 with seven enrichment columns — cancellations, prompt/response lengths, turn latency, code blocks, and edited/referenced files — backfilled from history still on disk.
• Claude Code and Codex parsers extract the new coach signals exactly; Copilot and Gemini contribute prompt/response lengths and fenced code where their logs carry text.
• Tool, project, sort, and other page-scoped controls now live in the primary header on every desktop page, switching to icon labels in compact windows.
• Sample mode (Shift+D) includes a demo coach payload, so the page can be explored without live data.

The Coach page

Report owns the composite grade and four practice-score cards, brings priority advice forward with direct links to each rule's evidence detail, and pairs equal-height KPI/sparkline Flow and Pace panels, including a visible single-day Flow state.

Findings is a filterable master/detail evidence explorer for every rule that fired.

AI Output leads with six outcome KPIs and a full-width period-aware bar and moving-average chart before the language, model, and project breakdown. Detail follows the selected period: 30-minute buckets for 24 Hours, hourly for 7 Days, full daily timelines for 30 Days and This Month, and monthly history for All Time.

Activity splits into three views:

• Work Hours: hourly pattern charts on a single-hue intensity ramp with a Less→More legend and exact counts in tooltips, plus a period trend.
• Calendar: a period-aware daily activity bar strip — turns per day with a 7-day average line, dimmed out-of-period context bars, a trailing ~2-month window for scoped periods and the trailing year for All Time — above a weekday/sessions/turns/spend day summary, a per-day session table (time range, project, tool, turns, cost, and a timeline track on a shared hour axis), and a fixed call inspector.
• Projects: ranked project activity cards with spend, estimated active hours (session blocks split at >15-minute gaps), calls, sessions, AI LoC, average per session, a tech-stack chip row derived from each project's AI code output, a work-pattern chip (weekday/weekend mix plus the dominant time of day), the project's most-edited files, and tool mix.

Session rows and the AI-Output project ranking use the same short project names as the rest of the app.

Richer archive and parsers

The calls table gains is_canceled, prompt_chars, response_chars, elapsed_ms, code_blocks_json, edited_files_json, and referenced_files_json. The v4 migration clears source_state once so history still on disk re-parses through the enriched parsers and backfills the new columns — it fills once and never clobbers. Rows whose source files are gone stay NULL and are excluded from rule denominators, so missing history cannot skew a grade.

Per tool:

• Claude Code: full prompt length, interrupts ([Request interrupted by user), turn latency, response length, code blocks from fences and Write/Edit/MultiEdit/NotebookEdit payloads, and edited/referenced files. User-line handling no longer records slash-command wrappers (<command-...>) or interrupt markers as the turn's prompt.
• Codex: user and agent messages, turn_aborted, and patch_apply_end changes (edited files plus added-line LoC). Session details now capture user prompts, which were previously empty for Codex rows.
• Copilot and Gemini: prompt/response lengths and fenced code where their logs carry text.
• Cursor: rows stay estimate-only by design.

Per-tool signal matrices are documented in docs/development/tools/<name>.md.

Upgrade notes

The local archive schema moves to version 4 on first open. The migration adds the enrichment columns and clears parser fingerprints once, so the first launch re-parses all history whose source files are still on disk — expect that first ingest to take longer than usual. The ingest cache format is now v2; older cache files are discarded and regenerated automatically.

Older tokenuse binaries refuse to open a version 4 archive. To downgrade after opening 1.1.0, delete archive.db and let the older version rebuild it from the source files still present on disk, backing up the database first if archive-only history matters.

Notes

• Local usage analysis remains local. The coach rules run entirely on your machine with no network or LLM calls; network use stays limited to confirmed Config-page downloads, opt-in quota sync, Windows/Linux update checks, and maintainer refresh or release paths.

Full Changelog: v1.0.0...v1.1.0

Full Changelog: v1.0.0...v1.1.0