Skip to content
Back to search

jevcore

TypeSafe Jev for DeepSeek Harness, the Model Context Protocol, and plain Node: typed judgments instead of prose, offline by default.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/perrylink--jevcore/badge.svg)](/listings/perrylink--jevcore)

HTML

<a href="/listings/perrylink--jevcore"><img src="/api/listings/perrylink--jevcore/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

As a DeepSeek Harness plugin

dsh plugin --profile <profile> add jevcore-dsh

Or from a checkout:

dsh plugin --profile <profile> add /absolute/path/to/jevcore/packages/dsh

`packages/dsh` imports `jevcore` by name, so a checkout also needs core resolvable from the profile (`add /absolute/path/to/jevcore/packages/core`).

Then confirm the row activated — the plugin list should show `jev` as `active`, not `failed` — and check the startup report in the log.

As an MCP server

For a host that speaks MCP, the same three tools are available over stdio:

npx -y jevcore-mcp

To wire it into DeepSeek Harness specifically, install a configuration-only bundle whose patch inserts the harness's own MCP client:

- insert:
    - id: jev-mcp
      name: '@deepseek-ai/dsh-mcp-client'
      config:
        serverName: jev
        transport: stdio
        command: npx
        args: ['-y', 'jevcore-mcp']
        env:
          TYPESAFE_API_KEY: '<the key>'
        failOnStartupError: true

The env map is not optional: DSH strips every credential-shaped name — anything containing KEY, PASSWORD, SECRET or TOKEN, in any case — from the environment it hands a spawned server, then merges this map back in afterwards. A key exported in your shell never arrives, and the server stays on the offline mock without reporting an error.

The provider is chosen from the environment:

| Variable | Effect | |---|---| | `TYPESAFE_API_KEY` | Selects the TypeSafe route when present | | `OPENROUTER_API_KEY` | Selects the OpenRouter route when present and no TypeSafe key is set | | `JEV_PROVIDER` | `mock`, `live`, or `openrouter` — overrides the heuristic above | | `TYPESAFE_MODEL` / `OPENROUTER_MODEL` | Model id for the selected route | | `TYPESAFE_BASE_URL` / `OPENROUTER_BASE_URL` | API root for the selected route |

With neither key it stays on the offline mock. Unlike the plugin, the MCP server resolves its credential once at startup, so a missing key with `JEV_PROVIDER=live` is a startup error rather than a per-call surprise.

Its egress report goes to **stderr**, never stdout — on a stdio transport stdout is the protocol channel, and a stray line there would corrupt the stream.

Going live

There are two routes to Jev. Both call the same models and both return the same typed answers; they differ in who holds your credential and whose servers see your state.

**TypeSafe directly** — use this if you have a key from [console.typesafe.ai](https://console.typesafe.ai/settings/keys):

- insert:
    - id: jev
      name: 'jevcore-dsh'
      config:
        provider: live
        apiKeyRef: TYPESAFE_API_KEY   # a reference, never the key
        model: jev-latest

**Through OpenRouter** — use this if a TypeSafe key is impractical and you already have an [OpenRouter](https://openrouter.ai) key. OpenRouter serves the System One models at the same `POST /v1/systemone` path TypeSafe does, one level below its own API root, so this is the documented route to Jev rather than an approximation of it:

- insert:
    - id: jev
      name: 'jevcore-dsh'
      config:
        provider: openrouter
        openRouterApiKeyRef: OPENROUTER_API_KEY
        model: jev-latest              # a bare `jev-*` id, or `typesafe/jev-1.13`

The route is reached by pointing the official `@typesafe-ai/sdk` at `https://openrouter.ai/api` with your OpenRouter key — OpenRouter's own documented integration, so there is no second client to keep in step.

Two things to know about the OpenRouter route:

- **Your state goes to OpenRouter, not to TypeSafe.** A different third party with different retention and logging. The startup report names the endpoint for exactly this reason — read it rather than inferring the destination from the provider's name. - **It returns a cost**, which TypeSafe's own route does not, so `usage.costUsd` is populated here and absent there.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 555 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # jevcore
  2. [![dshfind](https://dshfind.com/api/badge/PerryLink/jevcore?metric=downloads)](https://dshfind.com/plugins/PerryLink/jevcore?ref=badge)
  3. [![DSH Market](https://raw.githubusercontent.com/2BingLing/dsh-market/master/assets/readme/badge-listed-en.svg)](https://dsh.market/)
  4. [![OpenSSF Scorecard](https://api.securityscorecards.dev/projects/github.com/PerryLink/jevcore/badge)](https://api.securityscorecards.dev/projects/github.com/PerryLink/jevcore)
  5. [![jevcore MCP server](https://glama.ai/mcp/servers/PerryLink/jevcore/badges/score.svg)](https://glama.ai/mcp/servers/PerryLink/jevcore)
  6. TypeSafe [Jev](https://typesafe.ai) for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness)
  7. and any other MCP host.
  8. Jev is not a chat model. It answers typed questions — `noul` (yes/no), `choice`, `score` — and returns
  9. calibrated probabilities. It does not write prose, and asking it to is a category error. This project
  10. gives an agent exactly that surface, and nothing more.
  11. **Offline by default. Egress disclosed. Nothing default-on.**
  12. ---
  13. **📖 Ecosystem knowledge base** — measured data, not marketing: [plugin development guide · plugin-selection data · maintenance criteria](https://perrylink.github.io/dsh-plugin-guide/).
  14. <!-- star-cta -->
  15. ## ⭐ 如果它帮到了你
  16. 这个插件是 [DSH 插件家族](https://github.com/PerryLink)的一员(95 个,全部 Apache-2.0)。如果你在用,**给个 star** —— 它不会解锁任何功能,但会让下一个人在搜索里更容易找到它。
  17. *English:* part of a 95 plugin family for DeepSeek Harness. If it is useful, **a star helps the next person find it** — nothing is gated behind it.
  18. ## Three packages, one decision layer
  19. | Package | What it is | Use it when |
  20. |---|---|---|
  21. | [`jevcore`](packages/core) | The decisions. Imports nothing from DeepSeek Harness or Cordis. | You want Jev in a plain script, a service, or your own harness |
  22. | [`jevcore-dsh`](packages/dsh) | The DSH plugin: one service, three tools, two opt-in gates | You are running DeepSeek Harness |
  23. | [`jevcore-mcp`](packages/mcp) | The same three tools over MCP, with a stdio binary | Your host speaks MCP but is not DSH |
  24. The adapters are thin on purpose. `packages/dsh` is four files: it declares tool schemas and
  25. translates hook payloads. Everything decision-shaped — the primitives, the providers, the egress
  26. contract, the policy, the gates — lives in core, so a new adapter cannot drift from the guarantees
  27. the others make.
  28. One runtime requirement differs across the three: `jevcore-dsh` tracks the harness and needs
  29. Node `^22.19.0 || >=24.0.0`, while `jevcore` and `jevcore-mcp` need `>=20`.
  30. ---
  31. ## Why this exists
  32. Between 2026-09-17 and 09-20, nineteen plugins appeared that wire Jev into DSH. Auditing their source
  33. found a consistent pattern: the module labelled *guard*, *gate*, or *warden* was also the module
  34. shipping prompts, tool arguments, and file contents to a third party, and the README generally did not
  35. say so. Several were enabled by default. One gate could be reconfigured by the model it was guarding.
  36. This project is the same idea with those failure modes designed out:
  37. | Property | How it is guaranteed here |
  38. |---|---|
  39. | No network call unless you ask for one | Default provider is an offline mock; the live path needs both `provider: live` and a resolved credential |
  40. | Every transmission named before it happens | One startup log line per feature: `off` or `SENDS <feature> { fields }` |
  41. | Gates register nothing when disabled | Verified by test, not by policy — a disabled gate adds no event listener at all |
  42. | The model cannot widen its own constraints | No tool exposes gate configuration |
  43. | A judge that cannot answer never means "allow" | Undecided resolves through explicit config, defaulting to `ask` |
  44. ---
  45. ## The egress contract
  46. This is the part worth reading before installing.
  47. Transmission is decided per feature, and every feature defaults to off. The plugin prints its own
  48. contract at load:
  49. ```
  50. [jevcore] provider=mock endpoint=none egress=OFF (no network calls will be made; every answer is synthetic)
  51. [jevcore] ready - provider=mock - gates: safety=off context=off
  52. ```
  53. With `provider: live` and every feature enabled, the same report becomes explicit about what leaves:
  54. ```
  55. [jevcore] provider=live endpoint=https://api.typesafe.ai egress=ON
  56. [jevcore] SENDS tool:jev_ask { state<=16000c questions<=4000c }
  57. [jevcore] SENDS tool:jev_rank { state<=16000c questions<=4000c }
  58. [jevcore] SENDS tool:jev_check { state<=16000c questions<=4000c }
  59. [jevcore] SENDS gate:safety { state<=8000c questions<=2000c }
  60. [jevcore] SENDS gate:context { state<=6000c questions<=2000c }
  61. [jevcore] redaction is best-effort; it removes named fields and known secret shapes, and cannot
  62. recognise an unrecognised secret in free text
  63. ```
  64. | Feature | Off by default? | What it sends |
  65. |---|---|---|
  66. | `tool:jev_ask` | runs when the model calls it | the arguments the model passed, after redaction |
  67. | `tool:jev_rank` | runs when the model calls it | the query plus every candidate |
  68. | `tool:jev_check` | runs when the model calls it | the claim and its evidence |
  69. | `gate:safety` | **yes — explicit opt-in** | the tool name, its arguments, the workspace root |
  70. | `gate:context` | **yes — explicit opt-in** | a large tool result the agent just received |
  71. The tools transmit only when the model chooses to call them, which is visible in the transcript. The
  72. gates would run on every matching tool call, which is not, so they are opt-in.
  73. ### Redaction, and its honest limit
  74. Before anything is sent, `src/redact.ts` runs two passes: values under sensitive field names
  75. (`password`, `token`, `apiKey`, `authorization`, …) are replaced wholesale, and secret-shaped strings
  76. that survive into free text are pattern-matched (`Bearer …`, `sk-…`, `ts_live_…`, AWS/GitHub/Google
  77. key shapes, JWTs, private-key headers, credentials in connection strings).
  78. This is a mitigation, not a permission. A secret that sits under an unrecognised key name *and* does
  79. not match a known shape will pass through. If that possibility is unacceptable for your workload, do
  80. not enable the live provider.
  81. ---
  82. ## Install
  83. ### As a DeepSeek Harness plugin
  84. ```sh
  85. dsh plugin --profile <profile> add jevcore-dsh
  86. ```
  87. Or from a checkout:
  88. ```sh
  89. dsh plugin --profile <profile> add /absolute/path/to/jevcore/packages/dsh
  90. ```
  91. `packages/dsh` imports `jevcore` by name, so a checkout also needs core
  92. resolvable from the profile (`add /absolute/path/to/jevcore/packages/core`).
  93. Then confirm the row activated — the plugin list should show `jev` as `active`,
  94. not `failed` — and check the startup report in the log.
  95. ### As an MCP server
  96. For a host that speaks MCP, the same three tools are available over stdio:
  97. ```sh
  98. npx -y jevcore-mcp
  99. ```
  100. To wire it into DeepSeek Harness specifically, install a configuration-only bundle
  101. whose patch inserts the harness's own MCP client:
  102. ```yml
  103. - insert:
  104. - id: jev-mcp
  105. name: '@deepseek-ai/dsh-mcp-client'
  106. config:
  107. serverName: jev
  108. transport: stdio
  109. command: npx
  110. args: ['-y', 'jevcore-mcp']
  111. env:
  112. TYPESAFE_API_KEY: '<the key>'
  113. failOnStartupError: true
  114. ```
  115. The env map is not optional: DSH strips every credential-shaped name — anything
  116. containing KEY, PASSWORD, SECRET or TOKEN, in any case — from the environment it
  117. hands a spawned server, then merges this map back in afterwards. A key exported in
  118. your shell never arrives, and the server stays on the offline mock without
  119. reporting an error.
  120. The provider is chosen from the environment:
  121. | Variable | Effect |
  122. |---|---|
  123. | `TYPESAFE_API_KEY` | Selects the TypeSafe route when present |
  124. | `OPENROUTER_API_KEY` | Selects the OpenRouter route when present and no TypeSafe key is set |
  125. | `JEV_PROVIDER` | `mock`, `live`, or `openrouter` — overrides the heuristic above |
  126. | `TYPESAFE_MODEL` / `OPENROUTER_MODEL` | Model id for the selected route |
  127. | `TYPESAFE_BASE_URL` / `OPENROUTER_BASE_URL` | API root for the selected route |
  128. With neither key it stays on the offline mock. Unlike the plugin, the MCP server
  129. resolves its credential once at startup, so a missing key with
  130. `JEV_PROVIDER=live` is a startup error rather than a per-call surprise.
  131. Its egress report goes to **stderr**, never stdout — on a stdio transport stdout
  132. is the protocol channel, and a stray line there would corrupt the stream.
  133. ### Going live
  134. There are two routes to Jev. Both call the same models and both return the same
  135. typed answers; they differ in who holds your credential and whose servers see
  136. your state.
  137. **TypeSafe directly** — use this if you have a key from
  138. [console.typesafe.ai](https://console.typesafe.ai/settings/keys):
  139. ```yml
  140. - insert:
  141. - id: jev
  142. name: 'jevcore-dsh'
  143. config:
  144. provider: live
  145. apiKeyRef: TYPESAFE_API_KEY # a reference, never the key
  146. model: jev-latest
  147. ```
  148. **Through OpenRouter** — use this if a TypeSafe key is impractical and you
  149. already have an [OpenRouter](https://openrouter.ai) key. OpenRouter serves the
  150. System One models at the same `POST /v1/systemone` path TypeSafe does, one level
  151. below its own API root, so this is the documented route to Jev rather than an
  152. approximation of it:
  153. ```yml
  154. - insert:
  155. - id: jev
  156. name: 'jevcore-dsh'
  157. config:
  158. provider: openrouter
  159. openRouterApiKeyRef: OPENROUTER_API_KEY
  160. model: jev-latest # a bare `jev-*` id, or `typesafe/jev-1.13`
  161. ```
  162. The route is reached by pointing the official `@typesafe-ai/sdk` at
  163. `https://openrouter.ai/api` with your OpenRouter key — OpenRouter's own
  164. documented integration, so there is no second client to keep in step.
  165. Two things to know about the OpenRouter route:
  166. - **Your state goes to OpenRouter, not to TypeSafe.** A different third party
  167. with different retention and logging. The startup report names the endpoint for
  168. exactly this reason — read it rather than inferring the destination from the
  169. provider's name.
  170. - **It returns a cost**, which TypeSafe's own route does not, so `usage.costUsd`
  171. is populated here and absent there.
  172. The model id must be a System One one. `jev-latest` bare is the default and needs
  173. no prefix; `typesafe/` is accepted on a versioned id such as `typesafe/jev-1.13`,
  174. but not on the moving tag — `typesafe/jev-latest` is not accepted by the route.
  175. Any other id would be routed to a chat model, which answers with prose this plugin
  176. cannot interpret as a decision, so it is refused before the call rather than
  177. misread after it.
  178. Either way the credential is resolved through DSH's credential service first,
  179. then the environment variable of that name. It is read per call, so a key added
  180. while the process is running is picked up. It is never logged, never returned
  181. from a tool, and never written to configuration. Each route has its own reference
  182. (`apiKeyRef` and `openRouterApiKeyRef`) so the two cannot accidentally share a key.
  183. `@typesafe-ai/sdk` is the only optional dependency; the plugin
  184. loads and runs offline without either, and only needs the one for the route you
  185. choose.
  186. ---
  187. ## Configuration
  188. | Key | Default | Meaning |
  189. |---|---|---|
  190. | `provider` | `mock` | `mock` (offline, deterministic, synthetic), `live` (TypeSafe), or `openrouter` |
  191. | `apiKeyRef` | `TYPESAFE_API_KEY` | Credential reference for the `live` route |
  192. | `openRouterApiKeyRef` | `OPENROUTER_API_KEY` | Credential reference for the `openrouter` route |
  193. | `baseURL` | `https://api.typesafe.ai` | API root for the `live` route. Non-HTTPS is refused except on loopback |
  194. | `openRouterBaseURL` | `https://openrouter.ai/api` | API root for the `openrouter` route. Same rule |
  195. | `model` | `jev-latest` | Sent with every request. On the OpenRouter route the bare default works; `typesafe/` needs a versioned id such as `typesafe/jev-1.13`, and `typesafe/jev-latest` is not accepted |
  196. | `logLevel` | `warn` | `silent` \| `warn` \| `info` \| `debug` |
  197. | `minConfidence` | `0.7` | Below this, an answer is not acted on |
  198. | `minProbability` | `0.6` | Below this, a decision is not acted on |
  199. | `maxStateChars` | per feature | Replaces the `state` cap for every feature. `0` means "keep the declared cap" |
  200. | `gates.safety` | `false` | Judge tool calls before dispatch |
  201. | `gates.context` | `false` | Withhold large, uninformative tool results |
  202. The declared `state` caps are 16,000 characters for the three tools and 8,000 / 6,000 for the safety
  203. and context gates. `maxStateChars` replaces all of them, and the startup report shows the effective
  204. value rather than the declared one, so what it prints is what is enforced.
  205. Gates accept a bare boolean (`safety: false`) or an object with `onUndecided`: `ask` (default),
  206. `allow`, or `deny`.
  207. `ask` is a question, so it needs something to ask. A deployment that composes no approval service
  208. cannot escalate to a human, and DeepSeek Harness then refuses the call instead of running it: every
  209. tool the safety gate matches is refused, which an operator experiences as the plugin having broken
  210. shell commands. [docs/approval.md](docs/approval.md) has the mechanism, and the plugin README covers
  211. the deployment view.
  212. An unknown value is rejected at load with a message naming the key, rather than being silently
  213. ignored — a config typo should not quietly change the privacy posture.
  214. ---
  215. ## Using it
  216. ### From another plugin, with no model in the loop
  217. The service is the primary surface. This is the point of a decision model: a routing or gating
  218. decision should not cost a model round-trip.
  219. ```ts
  220. const jev = ctx.get('jev')
  221. const result = await jev.ask({
  222. feature: 'tool:jev_ask',
  223. state: { ticket: 'I was charged twice.' },
  224. questions: {
  225. urgent: { type: 'noul', instructions: 'Does this convey urgency?' },
  226. team: {
  227. type: 'choice',
  228. instructions: 'Which team should handle this?',
  229. criteria: { billing: 'Payments, invoices, refunds', technical: 'Bugs, outages' },
  230. },
  231. },
  232. })
  233. ```
  234. `result.answers` carries probabilities and confidence. Deciding what to do with them is your code's
  235. job — see `src/policy.ts` for a worked example with an explicit confidence floor, where an unsure
  236. answer produces `ask` rather than `allow`.
  237. ### From the model
  238. Three tools, deliberately few and orthogonal:
  239. - **`jev_ask`** — a batch of typed questions over one state.
  240. - **`jev_rank`** — score and sort candidates against one criterion, one question per candidate in a
  241. single round-trip. The probabilities are independent per-candidate judgments, not a distribution.
  242. - **`jev_check`** — does this evidence support this claim? Returns `supported`, `contradicted`,
  243. `conflicted`, `insufficient`, `undecided`, or `unknown`. Contradiction outranks support, because
  244. evidence that both supports and refutes is a conflict, not a weak yes.
  245. ### A bundled skill
  246. The plugin registers a skill, `typesafe-ai-dsh`, teaching an agent when a Jev
  247. judgment is the right tool and when it is a category error. It is registered
  248. through the skill registry rather than shipped as a directory for a provider to
  249. scan, so it needs no dependency on where a profile keeps its skills and it
  250. disappears cleanly when the plugin is removed.
  251. The body lives in `skills/typesafe-ai-dsh/SKILL.md` and is read at load rather
  252. than embedded, so the file a human edits is the file that ships. A test asserts
  253. the two cannot drift.
  254. The registry is declared as an **optional** dependency: a profile that does not
  255. compose the skill subsystem still gets the service and the three tools, with a
  256. warning that the skill was skipped.
  257. ### The mock provider
  258. With `provider: mock`, every answer is derived from a hash of the question and state, so tests can
  259. assert exact values and no socket is opened. Synthetic results are labelled as such in three places:
  260. `provider: "mock"`, a `mock/jev-synthetic` model name, and a `warning` field on the result. A mock
  261. that could be mistaken for a real judgment would be worse than no mock at all.
  262. ---
  263. ## Status
  264. Honest accounting of what has and has not been verified.
  265. **Verified**
  266. - 410 tests pass across three packages (314 core, 66 DSH, 30 MCP), with no network access and no
  267. `TYPESAFE_API_KEY`. CI clears the variable and expects the suite to pass anyway.
  268. - **The OpenRouter route is verified against the real API.** `pnpm --filter jevcore run
  269. probe:live` drives the provider and `pnpm --filter jevcore-mcp run smoke:live` drives the whole MCP
  270. surface — transport, tool schemas, service and provider — against real System One models
  271. (`typesafe/jev-1.13-20260917`). A three-primitive batch returned a `noul` at 0.91, a `choice` at
  272. 0.97, and a `score` of `1.05` on a four-level rubric, with usage reported as
  273. `{ inputTokens: 469, outputTokens: 68, costUsd: 0.000019698 }`. `jev_rank` ordered a credential
  274. runbook above a billing guide; `jev_check` returned `contradicted`. Both scripts need
  275. `OPENROUTER_API_KEY` and are excluded from CI.
  276. - **The payload shapes are checked against the vendors' own type definitions.**
  277. `test/vendor-conformance.test.ts` pins this project's question and answer types to both SDKs, so a
  278. drift in either direction is a compile error rather than a malformed request on the one route that
  279. costs money and transmits data. That check is what found `score.criteria` being sent as a keyed map
  280. when the API requires an ordered array — a defect the stubbed unit tests had passed over
  281. throughout. There is also a script that parsed our payloads against OpenRouter's published zod
  282. schemas for the `alpha/decisions` route; both the script and that route are gone, because the
  283. OpenRouter provider now uses the documented `/v1/systemone` path through the same client TypeSafe
  284. uses, which the conformance test already covers.
  285. - **The plugin activates in a running harness and its tools work.** The plugin row reports `active`;
  286. `jev_ask` returned `urgent=true (0.8307)` and `team=billing (0.5027)` against the mock in 1 ms, and
  287. `jev_check` returned `verdict="insufficient"` with its three probabilities. Every result carried
  288. `provider: "mock"` and zero token usage, so the default path made no network call.
  289. - **The bundled skill registers.** `typesafe-ai-dsh` appears in the session skill catalog.
  290. - The default path makes no network call: asserted by spying on `globalThis.fetch` while mounting the
  291. plugin and answering through the service, and again while assembling the MCP runtime.
  292. - A disabled gate registers **no** event listener, and a denied egress never reaches the provider.
  293. - `Config` satisfies the Standard Schema protocol Cordis requires before a plugin starts.
  294. - Every DSH API used here (`ctx.provide`, `ctx.effect`, `tools.register`, `defineTool`,
  295. `tools/pre-execute`, `tools/post-execute`, `credentials.resolve`) was checked against the installed
  296. runtime before use, and the payload types come from the installed declaration files.
  297. **Not verified, or known-broken**
  298. - **The TypeSafe route is exercised, but only lightly.** `packages/core/scripts/probe-live.mjs`
  299. (`pnpm --filter jevcore run probe:typesafe`) asks the real API three things: ten claim/evidence
  300. pairs whose verdict is known in advance, one question repeated six times, and a noul carrying a
  301. `criteria: {true, false}` boundary. Two separate runs agreed — supporting evidence scored 0.95,
  302. contradicting evidence 0.10, evidence silent about the claim 0.03, and the repeated question varied
  303. by 0.01 or less. The boundary was accepted. What remains untested is everything around the request
  304. rather than the request itself: quota, rate-limit and entitlement behaviour on a real account.
  305. - **`jev_ask`'s description carries a corrupted dash on the build that is currently running.** It
  306. reads `branches on —?routing` where an em dash followed by a space belongs. Cause: a UTF-8 round-trip
  307. early in development replaced the third byte of the em dash with `?`. It is fixed on disk — four
  308. occurrences, zero remaining, confirmed in both source and build output — but the running process
  309. loaded its module before the fix and cannot re-read it without a restart. The dash itself is
  310. cosmetic, but the staleness is not: the same process also lacks every fix made since it started,
  311. including the `band` field that keeps a 0.51 from being read as a settled yes.
  312. - The MCP server has been driven end to end by a real MCP client over stdio
  313. (`pnpm --filter jevcore-mcp run smoke`): handshake, tool discovery, three successful calls, and an
  314. error result for an invalid batch. It has not been driven by any other third-party host.
  315. - Gate behaviour on live traffic. The gates are tested against synthetic answers and real hook
  316. payload shapes, but no real tool call has been gated end to end.
  317. - The context gate cannot recover context already spent. It withholds a result from reaching the
  318. agent; it does not prune anything retroactively, and the README of any plugin claiming otherwise
  319. should be read sceptically.
  320. - No long-running or adversarial testing. Redaction is pattern-based and will miss an unrecognised
  321. secret shape.
  322. ---
  323. ## Design notes
  324. Three decisions that are easy to get wrong and expensive to get wrong:
  325. **A judge that cannot answer must not mean "allow".** If Jev is unreachable, or answers below the
  326. confidence floor, the gates resolve through `onUndecided`, which defaults to `ask`. The only way to
  327. get fail-open behaviour is to configure it. The context gate is the deliberate exception: it fails
  328. open unconditionally, because losing a real tool result to a phantom "irrelevant" verdict is worse
  329. than keeping an uninformative one.
  330. **Nothing the model can say changes the gate.** No tool exposes gate configuration, thresholds, or
  331. scope. A guard the guarded process can reconfigure is not a guard.
  332. **A probability is not a permission.** Jev returns numbers; `src/policy.ts` turns them into
  333. `allow`/`ask`/`deny` against locally configured thresholds. An answer that names a value outside the
  334. declared criteria is `invalid` and denies — the guarantee of a typed decision model is that it cannot
  335. return an undeclared value, so a violation means something upstream is wrong.
  336. ## Development
  337. ```sh
  338. pnpm install
  339. pnpm run check # typecheck + tests + build
  340. pnpm test # tests only
  341. ```
  342. No test requires a credential or a network connection, and CI enforces that by running with
  343. `TYPESAFE_API_KEY` cleared.
  344. ## PerryLink DSH Plugin Family
  345. This project is part of the DeepSeek Harness plugin family from [PerryLink](https://github.com/PerryLink). The roster stays collapsed below so every README keeps a short first screen; expand it to see every row and its status.
  346. <details>
  347. <summary>Expand all plugins</summary>
  348. | Plugin | Status |
  349. |---|---|
  350. | **[dsh-aqua-input-check](https://github.com/PerryLink/dsh-aqua-input-check)** | |
  351. | **[dsh-archive-check](https://github.com/PerryLink/dsh-archive-check)** | |
  352. | **[dsh-auto-review](https://github.com/PerryLink/dsh-auto-review)** | |
  353. | **[dsh-autotier](https://github.com/PerryLink/dsh-autotier)** | 🧊 FROZEN — see the repo README |
  354. | **[dsh-bid-ca-precheck](https://github.com/PerryLink/dsh-bid-ca-precheck)** | |
  355. | **[dsh-bid-qual-check](https://github.com/PerryLink/dsh-bid-qual-check)** | |
  356. | **[dsh-cert-mcp](https://github.com/PerryLink/dsh-cert-mcp)** | 🧊 FROZEN — see the repo README |
  357. | **[dsh-checkpoint-rewind](https://github.com/PerryLink/dsh-checkpoint-rewind)** | |
  358. | **[dsh-click](https://github.com/PerryLink/dsh-click)** | |
  359. | **[dsh-composer-history](https://github.com/PerryLink/dsh-composer-history)** | |
  360. | **[dsh-contract-stance](https://github.com/PerryLink/dsh-contract-stance)** | |
  361. | **[dsh-customs-doc-check](https://github.com/PerryLink/dsh-customs-doc-check)** | |
  362. | **[dsh-data-quality](https://github.com/PerryLink/dsh-data-quality)** | |
  363. | **[dsh-demurrage-ledger](https://github.com/PerryLink/dsh-demurrage-ledger)** | |
  364. | **[dsh-doublecheck](https://github.com/PerryLink/dsh-doublecheck)** | |
  365. | **[dsh-drill-script-check](https://github.com/PerryLink/dsh-drill-script-check)** | |
  366. | **[dsh-eia-guide-check](https://github.com/PerryLink/dsh-eia-guide-check)** | |
  367. | **[dsh-emergency-plan](https://github.com/PerryLink/dsh-emergency-plan)** | |
  368. | **[dsh-essay-rubric-check](https://github.com/PerryLink/dsh-essay-rubric-check)** | |
  369. | **[dsh-evidence-check](https://github.com/PerryLink/dsh-evidence-check)** | |
  370. | **[dsh-export-ctl-check](https://github.com/PerryLink/dsh-export-ctl-check)** | |
  371. | **[dsh-fast](https://github.com/PerryLink/dsh-fast)** | |
  372. | **[dsh-fmea-table-check](https://github.com/PerryLink/dsh-fmea-table-check)** | |
  373. | **[dsh-forecast-penalty](https://github.com/PerryLink/dsh-forecast-penalty)** | |
  374. | **[dsh-fund-research](https://github.com/PerryLink/dsh-fund-research)** | |
  375. | **[dsh-github](https://github.com/PerryLink/dsh-github)** | |
  376. | **[dsh-gongwen-flow-check](https://github.com/PerryLink/dsh-gongwen-flow-check)** | |
  377. | **[dsh-gongwen-word-check](https://github.com/PerryLink/dsh-gongwen-word-check)** | |
  378. | **[dsh-guard-plan-qc](https://github.com/PerryLink/dsh-guard-plan-qc)** | |
  379. | **[dsh-hazchem-check](https://github.com/PerryLink/dsh-hazchem-check)** | |
  380. | **[dsh-hazplan-check](https://github.com/PerryLink/dsh-hazplan-check)** | |
  381. | **[dsh-hidden-risk-map](https://github.com/PerryLink/dsh-hidden-risk-map)** | |
  382. | **[dsh-hs-classify](https://github.com/PerryLink/dsh-hs-classify)** | |
  383. | **[dsh-icd-rule-check](https://github.com/PerryLink/dsh-icd-rule-check)** | |
  384. | **[dsh-industry-research](https://github.com/PerryLink/dsh-industry-research)** | |
  385. | **[dsh-lawcite-adapter](https://github.com/PerryLink/dsh-lawcite-adapter)** | |
  386. | **[dsh-laya](https://github.com/PerryLink/dsh-laya)** | |
  387. | **[dsh-lc-doc-check](https://github.com/PerryLink/dsh-lc-doc-check)** | |
  388. | **[dsh-learning-gap-check](https://github.com/PerryLink/dsh-learning-gap-check)** | |
  389. | **[dsh-library](https://github.com/PerryLink/dsh-library)** | |
  390. | **[dsh-local-ai](https://github.com/PerryLink/dsh-local-ai)** | |
  391. | **[dsh-lsp-actions](https://github.com/PerryLink/dsh-lsp-actions)** | |
  392. | **[dsh-mask](https://github.com/PerryLink/dsh-mask)** | |
  393. | **[dsh-mcp-panel](https://github.com/PerryLink/dsh-mcp-panel)** | |
  394. | **[dsh-medrec-qc](https://github.com/PerryLink/dsh-medrec-qc)** | |
  395. | **[dsh-memento](https://github.com/PerryLink/dsh-memento)** | |
  396. | **[dsh-nsfc-form-check](https://github.com/PerryLink/dsh-nsfc-form-check)** | |
  397. | **[dsh-nurse-record-check](https://github.com/PerryLink/dsh-nurse-record-check)** | |
  398. | **[dsh-observe](https://github.com/PerryLink/dsh-observe)** | |
  399. | **[dsh-origin-rvc-check](https://github.com/PerryLink/dsh-origin-rvc-check)** | |
  400. | **[dsh-ota-review-check](https://github.com/PerryLink/dsh-ota-review-check)** | |
  401. | **[dsh-output-styles](https://github.com/PerryLink/dsh-output-styles)** | |
  402. | **[dsh-paper-doc-adapter](https://github.com/PerryLink/dsh-paper-doc-adapter)** | |
  403. | **[dsh-permission-rules](https://github.com/PerryLink/dsh-permission-rules)** | |
  404. | **[dsh-permit-report-check](https://github.com/PerryLink/dsh-permit-report-check)** | |
  405. | **[dsh-pipeline-check](https://github.com/PerryLink/dsh-pipeline-check)** | |
  406. | **[dsh-pleading-draft](https://github.com/PerryLink/dsh-pleading-draft)** | |
  407. | **[dsh-plugin-doctor](https://github.com/PerryLink/dsh-plugin-doctor)** | |
  408. | **[dsh-plugin-guide](https://github.com/PerryLink/dsh-plugin-guide)** | |
  409. | **[dsh-plugin-kit](https://github.com/PerryLink/dsh-plugin-kit)** | |
  410. | **[dsh-plugin-upgrade](https://github.com/PerryLink/dsh-plugin-upgrade)** | |
  411. | **[dsh-policy-brief-draft](https://github.com/PerryLink/dsh-policy-brief-draft)** | |
  412. | **[dsh-power-loss-split](https://github.com/PerryLink/dsh-power-loss-split)** | |
  413. | **[dsh-power-ticket-check](https://github.com/PerryLink/dsh-power-ticket-check)** | |
  414. | **[dsh-ppap-check](https://github.com/PerryLink/dsh-ppap-check)** | |
  415. | **[dsh-protest-deadline](https://github.com/PerryLink/dsh-protest-deadline)** | |
  416. | **[dsh-railway-window](https://github.com/PerryLink/dsh-railway-window)** | |
  417. | **[dsh-repair-order-qc](https://github.com/PerryLink/dsh-repair-order-qc)** | |
  418. | **[dsh-research-report](https://github.com/PerryLink/dsh-research-report)** | |
  419. | **[dsh-review-reply-check](https://github.com/PerryLink/dsh-review-reply-check)** | |
  420. | **[dsh-rulefile-check](https://github.com/PerryLink/dsh-rulefile-check)** | |
  421. | **[dsh-safety-brief-check](https://github.com/PerryLink/dsh-safety-brief-check)** | |
  422. | **[dsh-score](https://github.com/PerryLink/dsh-score)** | |
  423. | **[dsh-session-sync](https://github.com/PerryLink/dsh-session-sync)** | |
  424. | **[dsh-site-log-check](https://github.com/PerryLink/dsh-site-log-check)** | |
  425. | **[dsh-skill-pack-security](https://github.com/PerryLink/dsh-skill-pack-security)** | |
  426. | **[dsh-soe-decision-check](https://github.com/PerryLink/dsh-soe-decision-check)** | |
  427. | **[dsh-soilwater-check](https://github.com/PerryLink/dsh-soilwater-check)** | |
  428. | **[dsh-sop-sync-check](https://github.com/PerryLink/dsh-sop-sync-check)** | |
  429. | **[dsh-spc-gbt-adapter](https://github.com/PerryLink/dsh-spc-gbt-adapter)** | |
  430. | **[dsh-talk](https://github.com/PerryLink/dsh-talk)** | |
  431. | **[dsh-tender-extract](https://github.com/PerryLink/dsh-tender-extract)** | |
  432. | **[dsh-tender-matrix](https://github.com/PerryLink/dsh-tender-matrix)** | |
  433. | **[dsh-test-drive](https://github.com/PerryLink/dsh-test-drive)** | |
  434. | **[dsh-ticktick](https://github.com/PerryLink/dsh-ticktick)** | |
  435. | **[dsh-translate](https://github.com/PerryLink/dsh-translate)** | |
  436. | **[dsh-warranty-calc](https://github.com/PerryLink/dsh-warranty-calc)** | |
  437. | **[dsh-budget](https://github.com/PerryLink/dsh-budget)** | 🧊 FROZEN — see the repo README |
  438. | **[dsh-claude-move](https://github.com/PerryLink/dsh-claude-move)** | 🧊 FROZEN — see the repo README |
  439. | **[dsh-defend](https://github.com/PerryLink/dsh-defend)** | 🧊 FROZEN — see the repo README |
  440. | **[dsh-draw](https://github.com/PerryLink/dsh-draw)** | 🧊 FROZEN — see the repo README |
  441. | **[dsh-reach](https://github.com/PerryLink/dsh-reach)** | 🧊 FROZEN — see the repo README |
  442. | **[dsh-background-agents](https://github.com/PerryLink/dsh-background-agents)** | 🚫 RETIRED — see the repo README |
  443. | **[dsh-session-pin](https://github.com/PerryLink/dsh-session-pin)** | 🚫 RETIRED — see the repo README |
  444. | **[dsh-team-rooms](https://github.com/PerryLink/dsh-team-rooms)** | 🚫 RETIRED — see the repo README |
  445. </details>
  446. ## License
  447. [Apache License 2.0](LICENSE) © 2026 jevcore contributors
  448. TypeSafe, Jev, and System One are trademarks of TypeSafe AI. This project is an independent
  449. integration and is not affiliated with or endorsed by TypeSafe AI.
  450. <p align="center"><sub>工具链:<a href="https://github.com/PerryLink/dsh-plugin-guide">dsh-plugin-guide</a> · <a href="https://github.com/PerryLink/dsh-standards">dsh-standards</a> · <a href="https://github.com/PerryLink/dsh-plugin-start">dsh-plugin-start</a></sub></p>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.