dsh-deepseekbase

DeepSeek prices, plugin search, model specs and token trends — inside DeepSeek Harness. Four tools, one command.

dsh plugin --profile web add github:moldkey/dsh-deepseekbase

Open source on GitHub: github.com/moldkey/dsh-deepseekbase
MIT · zero runtime dependencies · single-file implementation · CI runs the four-stage check on every push

Four tools

Tool	What it answers
`deepseekbase_price`	What does DeepSeek cost, which provider is cheapest?
`deepseekbase_plugin`	Which plugin does X, how popular, how risky?
`deepseekbase_specs`	How big is the model, what context length, does it fit?
`deepseekbase_trend`	How much is DeepSeek actually used?

What it looks like in use

You: compare DeepSeek V4.1 Flash prices across providers

Tool returns:
  Official: off-peak in $0.15 / out $0.60 per 1M tokens
  Cheapest third party: Particle AI — in $0.025 / out $0.25
  6x spread on the same model

Why we built it

We run a free plugin test bench and found that what developers need is not another list — it is the answer to "does this install". Since we already compile that data, we put it where developers actually are: inside Harness.

Permissions: exactly one

Capability	Used
Outbound HTTPS	Yes — to `deepseekbase.com` only
File read/write	No
Credentials / env	No
Shell commands	No

Auditable in five minutes. The whole implementation is lib/index.js, one file, no runtime dependencies. No build artefact to trust.

We tested it with our own sandbox

Stage	Result
1 — install from GitHub	Pass (4.3s)
2 — appears in plugin list	Pass
3 — mounts in config tree	Pass
4 — profile boots and stays alive	Pass, no errors
GitHub Actions	Pass

Bugs we hit (documented)

Both are written up in the DSH plugin development guide:

Missing inject — calling ctx.systemPrompt without declaring it threw cannot get property "systemPrompt" without inject
pnpm workspace root — the first CI run failed on ERR_PNPM_ADDING_TO_ROOT

Plugin development guide · Plugin test results · MCP Server · Plugin directory