Skip to content
dsh-market Browse plugins GitHub 中文

cerebrixos-org/tuning-engines-cli#tuningengines-dsh-plugin

Exports metadata-only DSH turn, model, tool, approval, retry, and error events to Tuning Engines for governed traces, policy evaluation, cost analysis, and Work Session review, with a disk-backed retry queue.

Stars ★ 5 Category Notifications & Integrations Listed 2026-08-16

Install

Inside DeepSeek Harness, with dsh-market

dsh plugin --profile web add dshmarket

Or from the command line

dsh plugin --profile web add github:cerebrixos-org/tuning-engines-cli#path:/packages/tuningengines-dsh-plugin

Installing runs third-party code with your own permissions — it can read your files, use your credentials and reach the network. Review the source first, and pin a commit (github:owner/repo#sha) when you can.

README

Native DeepSeek Harness telemetry and tool governance for Tuning Engines. The plugin observes the Harness session event log, preserves turn/step/tool-call identity, and exports metadata-only traces through a durable local spool.

Install

export TE_API_KEY="your-tenant-token-or-inference-key"
dsh plugin --profile default add tuningengines-cli
dsh --profile default

Configuration

Environment variable Default Purpose
TE_API_KEY required Tenant API token; TE_INFERENCE_KEY is also accepted
TE_API_URL https://app.tuningengines.com TE control-plane URL
TE_GOVERNANCE_MODE observe off, observe, or fail-closed enforce
TE_TRACE_FLUSH_INTERVAL_MS 2000 Background flush interval
TE_TRACE_BATCH_SIZE 64 Maximum locally spooled records per flush
TE_TRACE_SPOOL_PATH ~/.tuningengines/dsh-trace-spool.jsonl Durable queue path

Data handling

The default and only v1 capture mode is metadata-only. Raw prompts, assistant content, tool arguments, tool results, credentials, and model reasoning are not exported. Tool arguments are represented by a SHA-256 fingerprint for stable correlation without content disclosure.

DeepSeek Harness remains the execution runtime. Tuning Engines supplies the governance, approvals, trace analytics, cost attribution, and review surfaces.

Content from the project README on GitHub ↗