Apache-2.0 DSH plugin by TencentCloud that streams GenAI traces to Tencent Cloud CLS with a five-layer entry/agent/step/chat/tool span model, Protobuf upload with no collector, configurable batching and retry, and a content-capture toggle.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-21
License
Apache-2.0
01What can it help you accomplish?
Get DSH GenAI traces into Tencent Cloud CLS without standing up a collector
One DSH turn becomes ENTRY → AGENT (invoke_agent) → STEP (react round_N) → CHAT / TOOL spans, uploaded as Protobuf directly from the plugin process
Teams already using Tencent Cloud CLS for logs who want agent traces next to them instead of running an OTLP pipeline
Control what content leaves the machine and how much
`captureContent` (or the standard `OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT` env var) turns prompt/response/tool payload capture off while keeping span structure, and `enabled: false` pauses collection without uninstalling
Operators who need traces for latency and error analysis but cannot ship source code or credentials to a log service
02How to install into DeepSeek Harness
Prerequisites
- DeepSeek Harness `>=0.1.0-rc.6 <0.2.0` (the README compatibility table caps it below 0.2.0)
- Node.js 18.0.0 or newer
- A CLS endpoint and topic id, plus either a SecretId/SecretKey pair or a numeric UIN
Installation steps
- 01
Install the DSH CLI globally if you have not: `npm install -g @deepseek-ai/dsh`.
$ npm install -g @deepseek-ai/dsh
- 02
Install into each profile you use: `dsh plugin --profile web add tencentcloud-agentobs-sdk-dsh` (also `headless` and `harness`).
$ dsh plugin --profile web add tencentcloud-agentobs-sdk-dsh
- 03
If pnpm v9+ blocks the protobufjs install script (`ERR_PNPM_IGNORED_BUILDS`), run `echo "enable-scripts=true" >> .npmrc` then `pnpm install` once inside the profile directory, e.g. `~/.dsh/profiles/web`.
- 04
Restart the DSH service — the README states an install or update only takes effect after a restart.
- 05
Configure it with env vars (`CLS_ENDPOINT`, `CLS_TOPIC_ID`, `CLS_SECRET_ID`/`CLS_SECRET_KEY` or `CLS_UIN`, `CLS_SERVICE_NAME`) or in `$DSH_HOME/profiles/<profile>/cordis.patch.yml`; explicit plugin config wins over env.
Verify the integration
Not specified by the author
Rollback
- `dsh plugin --profile web remove tencentcloud-agentobs-sdk-dsh` — repeat with `headless` / `harness` for the other profiles.
03DSH integration and capability boundaries
Native DSH observability plugin that hooks DSH session, agent-loop, LLM-stream and tool lifecycle events and uploads them to Tencent Cloud CLS through `tencentcloud-cls-sdk-js` — no OTLP collector or sidecar
Five-layer GenAI span model
DSH session events, agent-loop rounds, LLM streams and tool calls→ENTRY / AGENT (invoke_agent) / STEP (react round_N) / CHAT (chat model_name) / TOOL (execute_tool tool_name) spans in the Tencent Cloud AI-agent observability shape
spans are uploaded over Protobuf by `tencentcloud-cls-sdk-js` from the plugin itself, so no collector or sidecar is deployedBatching, queue and retry controls
`batchMaxSize` (32), `flushIntervalMs` (5000), `maxQueueSize` (2048), `retryTimes` (3), `debug`→Bounded batches flushed on an interval, with failed uploads retried
once the queue is full the oldest spans are dropped rather than blocking the agentContent capture and redaction scope
`captureContent` (default true), `contentMaxChars` (128000), or the env var `OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT`→Prompts, responses and tool arguments/results attached to spans, truncated per attribute, or omitted entirely
explicit plugin config in `cordis.patch.yml` takes precedence over environment variables
04Who is it for? When not to use it?
Good for
- Teams already using Tencent Cloud CLS for logs who want agent traces next to them instead of running an OTLP pipeline
- Operators who need traces for latency and error analysis but cannot ship source code or credentials to a log service
Not for
- The documented DSH range is `>=0.1.0-rc.6 <0.2.0`, with Node.js `>=18.0.0`; installing on a 0.2.x host is outside the supported matrix.
- Content capture is on by default, so source code, credentials and personal data can be sent to CLS. Turn it off with `captureContent: false` or `OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=false` and make sure CLS retention and access control match your policy.
- `maxBatchBytes` must stay below the SDK's 19 MB hard limit or the whole batch is rejected locally, and a full `maxQueueSize` queue drops the oldest spans.
05Compatibility, maintenance and safety notes
- The documented DSH range is `>=0.1.0-rc.6 <0.2.0`, with Node.js `>=18.0.0`; installing on a 0.2.x host is outside the supported matrix.
- Content capture is on by default, so source code, credentials and personal data can be sent to CLS. Turn it off with `captureContent: false` or `OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=false` and make sure CLS retention and access control match your policy.
- `maxBatchBytes` must stay below the SDK's 19 MB hard limit or the whole batch is rejected locally, and a full `maxQueueSize` queue drops the oldest spans.
Apache-2.0 · published by TencentCloud (latest release v0.0.1, 2026-08-18; last push 2026-08-26)
06Frequently asked questions
Do I need an OTLP collector?
No. The plugin calls `tencentcloud-cls-sdk-js` directly and uploads Protobuf batches, so there is no collector or sidecar to deploy — only an endpoint and topic id.
Which DSH versions are supported?
The README compatibility table lists `>=0.1.0-rc.6 <0.2.0` for DeepSeek Harness and `>=18.0.0` for Node.js.
Are my prompts and code sent to CLS?
By default yes — `captureContent` is `true`, so prompts, responses and tool arguments/results are attached to spans. Set `captureContent: false` or export `OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=false` to keep the span structure without payloads.
Install failed with ERR_PNPM_IGNORED_BUILDS — what now?
pnpm v9+ blocks dependency install scripts. In the affected profile directory (e.g. `~/.dsh/profiles/web`) append `enable-scripts=true` to `.npmrc` and run `pnpm install` once; later installs and updates do not need it again.
How do I remove it?
`dsh plugin --profile web remove tencentcloud-agentobs-sdk-dsh`, repeated for `headless` and `harness` if you installed there. Restart DSH so the removal takes effect.
07Related DSH workflows
dsh-univer-office
by dream-num
Give DeepSeek Harness a real office environment. Univer Office Plugin brings spreadsheets, docs, slides, canvases, relational tables, and more into one runtime — with connected data, validation, versioned changes, and isolated worktrees for multi-agent collaboration.
clearai-dsh
by clearailhc
Brings an evidence-checked epistemic loop to DSH: research sessions settle into a browsable domain ontology with evidence chains and support levels; contradictory readings are surfaced for human adjudication.
dsh-fish-memory
by yuuyuko-uu
Stores time-stamped, decaying per-session memories, recalls relevant items, and optionally distills imported local archives through the configured DSH model.
dsh-trading
by zhu1090093659
Agent-native trading terminal built on DeepSeek Harness. Crypto, US, CN and HK in one three-column GUI, 19+ hot-swappable connectors, dry-run by default with human approval on every live order. BYOK, no data redistribution.
08Data and sources
`tencentcloud-agentobs-sdk-dsh` 是一个 DeepSeek Harness (DSH) 可观测插件,直接将 GenAI trace 数据上报到腾讯云日志服务 (CLS)。
它观察 DSH 原生的 session、agent loop、LLM stream 和 tool 生命周期,将其转换为腾讯云AI Agent可观测规范的 5 层 span 层级模型(entry → agent → step → chat →…
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-21. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.
