# DeepSeek Harness Usage Statistics

[简体中文](README.md) · [Downloads](https://github.com/Missher12/Missher-DSH-Usage-Statistics/releases) · [Desktop application](https://github.com/Missher12/Missher-DeepseekHarness-Desktop)

Local session and Token statistics with daily and hourly activity charts for DeepSeek Harness.

An independently installable, removable Harness Bundle named `@missher/dsh-usage-statistics`. Open **Settings → Usage Statistics** after enabling it. Reading existing records does not require an API key. The current source/package version is **0.2.1**; see Releases for versions available to download.

## Features

- Total Tokens, peak daily Tokens, longest active session duration, and current/longest activity streaks.
- A 53×7 activity grid covering the last 53 weeks, with daily, weekly and cumulative views and exact-value tooltips.
- A frameless chart of today's 24 clock hours. Hover to see the time range and exact Token count, navigate hours with arrow keys, or refresh manually.
- A shared chart color preference: follow the theme, choose one of five presets, or select a custom color.
- Input cache hit rate, most-used model and reasoning effort, and tool/skill rankings; unreadable sessions and missing usage reports are disclosed.
- English/Chinese, host light/dark themes, a cached snapshot and retry controls.

## Host and platform support

Requires the DSH **0.2.0-rc.2** `sessionPersistence`, `storageDomain`, Typert Remote, `settings.section` and native UI primitives APIs. The plugin does not call Missher-only extension APIs or require a separate compatibility plugin.

Existing installation, Loader, RPC and controlled-browser evidence comes from **Intel macOS with the rc.2-based Missher SDK/Desktop environment**. Version 0.2.1 retains the three validated runtime entry points from `0.2.1-local.9`; this release updates packaging and documentation only. **The full unmodified official app, Windows, Linux, Apple Silicon and official alpha.1 have not been independently validated with this version.** Desktop application tests are not treated as plugin tests. See [VALIDATION.md](VALIDATION.md) for evidence boundaries.

DSH peer ranges remain `*`, so a host is not rejected by its version number alone. This is not a promise of compatibility with every version. Missing required APIs should remain ordinary loading errors; no compatibility bypass is needed.

## Install and use

1. Download `missher-dsh-usage-statistics-<version>.tgz` from [Releases](https://github.com/Missher12/Missher-DSH-Usage-Statistics/releases) and verify it against that release's `SHA256SUMS`. GitHub's automatic Source code archives are not the plugin installer.
2. In Desktop, open **Plugins → Add plugin → Package name or address**. Enter the downloaded archive's absolute path, or its public download URL. The archive includes all runtime entry points; users do not need to build it or provide an SDK.
3. Enable/reload when prompted, then open **Settings → Usage Statistics**. Existing records are aggregated; an empty profile shows an empty state.
4. Turn off the plugin's enable switch in plugin management to disable it. Turn it back on to restore it. If the host is not using hot reload, restart normally when prompted.
5. Uninstall `@missher/dsh-usage-statistics` from the same plugin page. Its settings section, service and styles are removed with the plugin lifecycle. Disabling/uninstalling does not delete sessions, model settings or credentials. Rebuildable caches and color preferences remain.

Back up the active profile before updating. Keep the same package name and avoid installing an older renamed copy alongside it.

For an initialized **Web/CLI profile** named `web`:

```sh
dsh plugin --profile web add /absolute/path/to/missher-dsh-usage-statistics-0.2.1.tgz
dsh --profile web --no-open
# After stopping the Web profile:
dsh plugin --profile web remove @missher/dsh-usage-statistics
```

Use the Desktop plugin page for the application's own profile. The public repository also contains built `lib` entry points and a Bundle manifest, so a Git source can be packaged without a build script. The versioned release archive is the reproducible distribution tested here; this packaging revision does not claim a new Git-install test.

“Local plugin directory” means an already-built directory with resolvable runtime dependencies. Unpacking an archive alone does not prepare peers for a linked-directory installation. Development SDK links are local to the developer's machine; distribute the `.tgz`, not those links. See [INSTALL.md](INSTALL.md).

## Data and accounting

Only readable sessions in the current Harness data directory are included. This is recorded usage, not a provider bill, account quota check or estimate, and it does not aggregate other computers.

- Tokens include uncached input, output, cache reads and cache writes. Reasoning Tokens are not added again on top of output. Missing usage is not estimated. Reported failed attempts and retries count; duplicate settlements within an attempt and inherited session prefixes are not double-counted.
- Daily and hourly charts use the same time zone, defaulting to the system zone. Accepted usage events are assigned to the hour when they were recorded; requests spanning hours are not spread evenly. Repeated daylight-saving hours are merged and skipped hours remain zero. Opening the page or refreshing reloads data; an older cached date is not labeled “today.”
- Cache hit rate is cache-read Tokens divided by all input Tokens. Session duration sums completed turns' active durations, excluding idle time between turns.
- Model and reasoning-effort values describe recorded history, not current defaults or declared model capabilities. The combined tool/skill ranking shows the top five entries.

The Host reads through the session persistence service and writes only the plugin's rebuildable `missher_usage_statistics` cache (domain 1, row format 4). The browser stores a snapshot and color preference. The plugin does not scan private log formats directly, edit original sessions, or repair/migrate historical logs. Unreadable records are reported as omitted.

It registers no model tools, adds no prompts, calls no model and uploads no statistics. It does not provide cross-device sync, cost forecasts or current-session context management. Large histories may take longer to scan; the default refresh budget is 12 seconds. Advanced options are `timeZone` (an IANA time zone) and `refreshTimeoutMs` (10–60000 milliseconds).

## Development

End users install the built archive. Source development requires Node.js, pnpm 11.7.0 and a built DSH 0.2.0-rc.2 SDK:

```sh
node scripts/link-harness.mjs /absolute/path/to/built-harness
pnpm install --frozen-lockfile
pnpm typecheck
pnpm test --maxWorkers=1
pnpm build
pnpm pack:bundle
```

The `harness-sdk` link, development dependencies and local paths are excluded from the distribution. Use an isolated copy if the working checkout is linked to a running application. `DSH_SOURCE_DIR=/absolute/path/to/built-harness node scripts/verify-profile.mjs` creates its own acceptance profile; it is not a command for testing against daily user data.

## Attribution and support

MIT licensed. The initial implementation was extracted from Desktop 0.5.10 / Harness 0.1.5-rc.2 at revision `ca0085cabd778685b83c79ff40e49d637014b28f`. This plugin maintains later API adaptations, hourly activity and chart colors. [provenance.json](provenance.json) lists original files and hashes.

The archive retains the [project MIT license](LICENSE), [DeepSeek MIT notice](licenses/deepseek-harness-MIT.txt) and [Zod MIT notice](licenses/zod-MIT.txt). The browser entry bundles Zod 4.4.3; Cordis, Schemastery and DSH services are supplied by host dependencies.

Report the host/plugin versions, platform, reproduction steps and redacted errors in [Issues](https://github.com/Missher12/Missher-DSH-Usage-Statistics/issues). Do not attach credentials or private session content.