traceme-insights

name	traceme:insights
description	Multi-day trend analysis — token consumption, session time, and skill usage rankings across projects

TraceMe Insights

Analyze token trends, time distribution, and skill usage across your projects over multiple days.

Commands

node "${CLAUDE_PLUGIN_ROOT}/scripts/traceme-cli.mjs" insights                 # Last 7 days (cross-device)
node "${CLAUDE_PLUGIN_ROOT}/scripts/traceme-cli.mjs" insights --day           # Today only
node "${CLAUDE_PLUGIN_ROOT}/scripts/traceme-cli.mjs" insights --month         # Last 30 days
node "${CLAUDE_PLUGIN_ROOT}/scripts/traceme-cli.mjs" insights --days 14       # Explicit N days
node "${CLAUDE_PLUGIN_ROOT}/scripts/traceme-cli.mjs" insights --local         # Local DB only
node "${CLAUDE_PLUGIN_ROOT}/scripts/traceme-cli.mjs" insights --project NAME  # Filter to project

Output Sections

1. Quick Stats — days analyzed, projects, sessions, prompts, tokens, cost, session time, skills
2. Token Consumption by Project — per-day per-project token table with totals
3. Time Consumption by Project — session count, total and average duration (zombie sessions excluded)
4. Skill Usage Rankings — ranked table with bar chart; per-project breakdown when multiple projects
5. Model Usage — per-model calls, tokens, cost (cross-device when synced; aggregated globally, so omitted under --project)

Visual Dashboard

For a fully interactive, graphical view, generate the HTML dashboard:

node "${CLAUDE_PLUGIN_ROOT}/scripts/traceme-cli.mjs" dashboard            # Generate & open in browser
node "${CLAUDE_PLUGIN_ROOT}/scripts/traceme-cli.mjs" dashboard --no-open  # Write file only

It writes ~/.claude/traceme/dashboard.html and opens it. The page embeds the last 90 days of data and renders with Apache ECharts (loaded from a CDN — needs internet on first open). Everything is filtered in-browser, no CLI re-run: pick the date range, filter by one or more projects and devices (all devices vs. a single device, when sync is set up), switch grouping by model / project / device / category, toggle the calendar intensity between billable tokens and cost, and toggle the cache_read layer.

Cross-device: the token/cost/session/calendar/trend views combine the local live DB with each foreign device's synced snapshots. Only tokens/cost/session counts are synced — per-model, tool-category, and skill breakdowns plus "Elapsed" are local-device only (those panels carry a note and the basis/cache_read toggles appear only in single-local-device mode).

Honesty notes baked into the view: the calendar/trend default to billable tokens (input+output+cache_creation) — re-read cache is excluded unless toggled, so idle big-context sessions don't look huge; the tool-category chart keeps subagent (actual tokens) separate from MCP/plugin/builtin (a coarse ≈ result-bytes estimate — not comparable, no shared %); "Elapsed" is gross wall-clock (includes idle), with sessions bucketed by their start day. Run traceme rescan --all once to backfill older sessions, then re-run dashboard.

Data Source

All sections — tokens, session duration, skill usage, and model usage — use the cross-device merged sync snapshot by default (all your machines). Foreign snapshots pushed before a field existed (active_min, skill_usage) contribute 0 for it until that device re-pushes (traceme sync push --all). Pass --local to use this device's data throughout. Tool-category breakdowns remain local-only (not synced).

Privacy

No prompt text is included in any section of the insights output.