# Usage Dashboard

> Show token usage and estimated cost across local Claude Code instances. Use when the user asks to see their Claude Code usage, token usage, cost, spend, or a usage dashboard — especially across multiple accounts/config dirs (~/.claude and ~/.claude-myfinancial).

- Skill: `rohitguta2432/usage-dashboard` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add rohitguta2432/usage-dashboard`
- Raw SKILL.md: https://api.skillmd.com/api/skills/rohitguta2432/usage-dashboard/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: rohitguta2432 (https://skillmd.com/u/rohitguta2432)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/rohitguta2432/usage-dashboard

---


# Usage Dashboard

Reads every `~/.claude*/projects/**/*.jsonl` transcript, sums input/output/cache
tokens + estimated cost, groups by instance, model, and day, then opens an HTML
dashboard. Each `~/.claude*` config dir is treated as one instance (label =
dir suffix; `~/.claude` → `personal`, `~/.claude-myfinancial` → `myfinancial`).

## Run

```bash
python3 ~/.claude/skills/usage-dashboard/dashboard.py
```

Writes `claude_usage.html` to the temp dir and opens it in the browser.

## Daily Obsidian ledger

`daily_obsidian.py` reuses the same collector to write the usage into the
Obsidian vault (`$VAULT_DIR`, default `~/Documents/wiki`):

```bash
python3 ~/.claude/skills/usage-dashboard/daily_obsidian.py
```

- One note per day at `wiki/usage/<YYYY-MM-DD>.md` — frontmatter
  (`cost`/`tokens`/`turns`, Dataview-queryable) + tables: by instance, by
  project, by model, token composition.
- `wiki/usage/index.md` — rolling master table (newest day first) + all-time totals.
- Idempotent: regenerates every run, self-heals missed days.
- Scheduled daily at 22:05 by `com.rohit.usage-ledger.plist` (installed in
  `~/Library/LaunchAgents/`), just after the `claude-to-obsidian` chat sync (22:00).
- Self-check: `python3 daily_obsidian.py --selfcheck`.

## Scheduled-run attribution

Both the HTML dashboard and the Obsidian note split usage by **what triggered
the run**: `Interactive` (you typed it) vs each scheduled launchd job by name
(`daily-financial-blog-post`, `daily-code-feature-add`, `daily-self-improve`, …).

- HTML: the **Schedule** grouping button (next to Project/Model/Instance);
  click a row to drill into its tokens/models/turns.
- Obsidian: a top-level **By schedule** table + a per-account one in each note.
- Detection (in `dashboard.collect()`): a launchd job runs headless with its
  `SKILL.md` as the prompt, so the session's first user message starts with that
  skill's YAML frontmatter (`---\nname: <task>`). `TASK_RE` extracts the name;
  anything else is `interactive`. No plist parsing, no API calls.

## Notes

- Prices are list-rate estimates in the `PRICING` dict at the top of
  `dashboard.py`. They drift — edit them to match your actual plan.
- Self-check: `python3 dashboard.py --selfcheck` (verifies cost math).
- Adds a new instance automatically when a new `~/.claude*` dir appears.
- Instance label map (`personal` → `rohitgupta`) lives in `NAMES` in
  `daily_obsidian.py` and `NAME` in `dashboard.py` — edit to taste.
- Lazy alternative for a single instance: `npx ccusage`.