Mock Harness

The mock harness replaces the real CLI spawn with a scripted event stream. Use it for tests, CI, and demos where real credentials or network access are undesirable.

Enabling

Two equivalent switches:

adapters run claude --use-adapters-harness-mock --prompt "test"

or globally:

export USE_MOCK_HARNESS=1
adapters run claude --prompt "test"

Programmatically:

await client.run({
  agent: 'claude',
  prompt: 'test',
  useMockHarness: true,
});

Scenarios

The mock emits a deterministic sequence of AgentEvents chosen by scenario name:

Pick one with --scenario:

adapters run claude --use-adapters-harness-mock --scenario tool-call --prompt "x"

If you invoke the package binary directly, the same agent scoping is available there too:

adapters-harness-mock --agent claude --list
adapters-harness-mock --agent claude --scenario tool-call

Interactive approval scenarios under packages/adapters/harness-mock now model real gating semantics rather than timer-only replay:

• interactive:yolo auto-approves and then emits the post-approval output.
• interactive:prompt waits for stdin before it emits the post-approval output and exits.
• interactive:deny auto-denies, emits the denial path, and exits non-zero.
• interactive:timeout waits for approval until the configured timeout, emits a timeout error, and exits non-zero.

Why use it

• **CI**: No API keys, no flakiness.
• **Adapter tests**: Validate parseEvent without invoking the real CLI.
• **Integration tests**: Exercise hooks, session-save paths, and the event stream in isolation.

See Harness & Mock for the full event script format.