Skip to content

Softprobe Agent QA

Capture real agent sessions. Review steps and Findings. Improve agents with production evidence.

Agent QA connects coding agents (starting with OpenCode) to Softprobe so every turn, model call, and tool use lands as a Session you can inspect in Explorer.

What Agent QA does

  • Install with one OpenCode prompt — paste from Explorer; the agent installs @softprobe/opencode-plugin and credentials
  • Capture full Sessions — user turns, generations, tools, retries, and errors
  • Review in Explorer — Sessions, Steps, and Findings for production behavior
  • Optional Slack — attach a channel after capture is verified

What Agent QA is not

ProductUse when…
Agent EvaluationYou run offline/CI suites, evaluators, and release gates
TestingJava record/replay regression with the JVM agent
PlatformIstio/SESSIFY business-flow observability

Agent QA is the online capture and review path for AI agents. Evaluation consumes evidence; Testing and Platform cover non-LLM services.

Start here

  1. Quick start — connect from Explorer
  2. OpenCode install — plugin, credentials, troubleshooting
  3. Concepts — Session and observation shape

Zero code changes · Full-context visibility · Cost optimization