AI Agent Hub
Back to plugins
🧩

dsh-observability

admin-security Updated 2026.08.20

Run the following command in DeepSeek Harness:

dsh plugin install xingyingyuzhui/dsh-observability

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install xingyingyuzhui/dsh-observability in the DeepSeek Harness terminal to install this plugin; see https://github.com/xingyingyuzhui/dsh-observability for details

About this plugin

When several plugins cooperate inside a governance suite, a single operation can span multiple modules, yet there is no unified timeline to reconstruct what actually happened. dsh-observability closes that gap: it ships a shared diagnostic-logging protocol (observe.mjs) so sibling plugins write structured records through the same path, and then renders those records on a Runtime Diagnostics page where you can filter by error code or traceId to see the full sequence at a glance.

The design is deliberately restrained. It logs only the end-state of an operation—succeeded, rejected, failed, or degraded—rather than every internal function call. A strict field whitelist keeps sensitive material out of the log: prompt text, memory and identity content, tool arguments, and tokens are never written. Records land in $DSH_HOME/agent-observe/diag.jsonl, and if the file is unwritable the writer falls back to stderr without blocking the main flow. Security auditing remains the responsibility of dsh-agent-gate, keeping the two concerns cleanly separated.

It is a practical fit for operations and security teams who run the dsh suite and need to pinpoint which step in a multi-plugin chain went wrong, without pulling in a heavyweight APM stack or leaking user data into log files.

Use Cases

  • Reconstruct the full operation timeline by traceId when a multi-plugin chain fails
  • Audit operation end-states rather than every internal function call
  • Fall back to stderr when the log file is unwritable, without blocking the main flow

Best For

  • Ops teams running the dsh suite who need to pinpoint errors across multi-plugin chains
  • Developers who care about log safety and do not want prompt text or tokens on disk
  • Security auditors using the governance suite who need a unified diagnostics view