dsh-llm-trace

DeepSeek Harness plugin that captures the raw HTTP request and response bodies of model provider calls and shows them in a Conversation View tab and on a loopback viewer page. 适合需要调试、审计或查看模型接口原始通信报文的 DSH 开发者和高级用户。

패키지

dsh-llm-trace

호환성

미검증

Cordis peer 범위

^4.0.2

버전

0.1.1

라이선스

Apache-2.0

최근 업데이트

2026. 9. 21.

Configuration

Every field is optional. The tab and the page share one config; the tab's mount path is the constant BASE at the top of lib/client.js.

Field	Default	Meaning
`path`	`/llm-trace`	Mount path; absolute, no trailing slash. Changing it also requires editing `BASE` in `lib/client.js`.
`match`	`['/chat/completions', '/v1/messages', '/v1/responses', '/v1/completions', '/responses']`	URL substrings selecting a provider call, matched case-insensitively.
`matchAll`	`false`	Capture every request regardless of `match`.
`redactHeaders`	`['authorization', 'api-key', 'x-api-key', 'proxy-authorization', 'cookie', 'set-cookie']`	Header names stored as ``. Set `[]` to keep them verbatim.
`maxRequestBodyBytes`	8 MiB	Retained request-body prefix.
`maxResponseBodyBytes`	32 MiB	Retained response-body prefix.
`maxRetained`	500	Exchanges retained before the oldest is evicted.
`maxJournalBytes`	256 MiB	Total retained body bytes across all exchanges.

Set them from a profile patch:

- insert:
    - id: llm-trace
      name: 'dsh-llm-trace'
      config:
        match: ['model.example.com']
        maxRetained: 2000

dsh-llm-trace

설치

문서

Configuration