workbench__test_network

Tests the drafted, unsaved network once, after you approve the run.

Behavior

• Starts at an agent node with a message, or runs a plan or tool node with an input.
• Every agent runs on its own model and tools run for real. Nothing is saved. A test stops after 90 seconds.
• Needs every node drafted without problems, and returns which nodes ran and the answer.
• Before anything runs, collects the secrets every tool the test would call needs, drafted or existing: each ${NAME} with no default in a tool’s env. Each resolves from the environment, then the project’s stored value, then the global one. When any is missing, nothing runs and the result is status: missing_secrets, listing each missing name with the tools that need it and its description; the network then waits on those secrets until a test runs.

Available to	Effect
Workbench agents only	Has side effects

Input

entry · string · required

The agent or plan node the test starts at

input · object

Plans: the plan’s input

message · string

Agents: the user message to send

Output

answer · string

passed · boolean

ran · string[]

Output

{
  "answer": "…",
  "passed": true,
  "ran": [
    "release_desk",
    "plan__release_digest"
  ]
}