Introduction¶
When working with DeepSeek Harness (DSH) and the Maestro plugin stack, developers need to manage tool calls, cost analysis, and system health status. Although general-purpose solutions exist, local, lightweight tools tailored to Maestro-specific workflows are usually preferred. dsh-maestro-observe provides an SQLite-based local storage solution to address these concerns.
Plugin Information¶
- Name: dsh-maestro-observe
- Core Value: Local-first observability for the Maestro plugin stack: trace, cost, and health capture + client dashboard.
- Maintainer: ddtcorex
- Category: Client
- Stars: 1
- License: MIT
Installation and Activation¶
Install it using the official command:
dsh plugin add @ddtcorex/dsh-maestro-observe
Core Capabilities¶
The plugin records signals (traces, health, cost) into local SQLite storage and exposes them through command-line tools and RPC endpoints.
- Local storage: Signals are stored in
~/.dsh/dsh-maestro-observe/observe.sqlite. - Trace: The
tracecommand supportslimit,sessionId,tool,kind, andsinceparameters. - Cost: The
costcommand supports queries by day or session scope and supportsgroupBy. - Errors: The
errorscommand groups errors by tool. - Latency: The
latencycommand providescount,p50,p95, andp99metrics. - Health: The
healthcommand reports plugin, tool counts, and status. - Budget: The
budgetcommand is used to set and check token limits. - Configuration: The
configcommand is used to get and set operational parameters. - Status RPC: The
/dsh-maestro-observeendpoint provides uptime, version, ring size, and row count.
Optional features include Telegram digests and peak alerts via maestroNotifier.
Typical Usage¶
trace { limit?, sessionId?, tool?, kind?, since? }cost { scope: 'day'|'session', sessionId?, day?, groupBy?: 'tool'|'session' }budget { action: 'set'|'check', scope, key, limit_tokens? }config { action: 'get'|'set', key, value? }errors { tool?, since? }latency { tool?, since? }health {}
Notes¶
- Runtime: The plugin runs with the permissions of the current DSH process. Check the source code and license before installing.
- Dependencies: Requires Node.js ^22.19.0 || >=24.0.0 and
@deepseek-ai/cordis(~4.0.4). - UI: There is no built-in client UI; rendering occurs in
dsh-maestro-dashboard. - Privacy: Sensitive information in trace details is redacted before persistence.
- Telegram: Digests are scheduled in local time (HH:MM). To send alerts,
@ddtcorex/dsh-maestro-notifieris required; otherwise, a warning is logged.
Summary¶
This plugin fills the gap in local debugging and observability requirements for the Maestro plugin stack. For more details and the source code, see the community directory or GitHub.