dsh-team-reporter

A DeepSeek Harness (dsh) plugin that reports this machine's real token usage to a team dashboard you run yourself, so a team can see its own usage and cost in one place.

What it does

• Collects one record per model call, from the host's session/event stream — input_tokens, cache_read_tokens, output_tokens, the model name, and the project directory. Reasoning tokens are already included in output_tokens and are not counted separately.
• Queues records on disk, so usage survives a restart or a network outage and is replayed later. Records are idempotent, so a replay cannot double-count.
• Sends only token counts and cost. It never sends prompts, replies, file contents, or credentials.
• Syncs the skills your team publishes: the server is the authority, the plugin pulls the whole library every 30 minutes and swaps it in atomically. A failed sync never removes what is already on the machine. You can also write your own skill and submit it for review from the panel.

Install

Download the tarball and point the plugin page at the local file — this is the only route that works without npm, without GitHub, and without a terminal:

1. Get dsh-team-reporter-0.2.0.tgz from your team's download page (or from the Releases tab here).
2. DSH → sidebar 插件 → 添加插件 → paste the file's absolute path → Install.
3. Restart DSH.

# CLI equivalent (the path must be absolute; a relative path is rejected by pnpm)
dsh plugin --profile desktop add "C:\Users\you\Downloads\dsh-team-reporter-0.2.0.tgz"

Neither of these works, and both were tested:

add https://github.com/CHENKEYI-MAKER/dsh-team-reporter.git does work, but it needs GitHub access.

Restart DSH. A grey dot appears in the bottom-right corner — open it, enter your team server URL and the enrollment code your administrator gave you, and click 绑定 (Bind). The dot turns green and reporting starts.

A code is single-use and belongs to one person. Reinstalling or moving to a new machine needs a new code from your administrator.

Skills

The plugin keeps two directories strictly apart:

Put your own skills in my-skills/, then submit them for review from the team panel (团队版 → 我写的技能 → 提交审核). Anything that looks like a credential, a public IP, or a local absolute path is rejected at submission time and cannot be approved by an administrator either.

Configuration

The plugin learns everything else from the server's /api/v1/config endpoint — the price table used for local cost estimates and the flush interval. Nothing needs to be configured locally.

Requirements

• Node.js >= 22.5.0
• A team server you or your administrator run (this plugin only talks to the URL you enter)

Privacy

Token counts and a hash of the session id are the only things that leave the machine. The workspace field is the project directory's name, not its full path.