The vidai-mock skill covers VidaiMock, a
local mock server for LLM provider APIs. A single process serves
provider-shaped OpenAI, Anthropic, Gemini, Bedrock, and compatible endpoints on
port 8100 with no API key and no network access, because the bundled providers
and templates are compiled into the binary.
Use it for LLM integration tests that must exercise streaming, tool calls, agentic loops, and failure handling without spending provider tokens. It reproduces the parts of a real provider that tests depend on: time to first token and token pacing, each provider's own streaming frame format, and chaos injection that returns provider-shaped error envelopes so retry and fallback logic engages the way it does in production.
Start it with vidaimock --host 127.0.0.1, confirm GET /health reports
{"status":"ok"}, then point the SDK under test at http://localhost:8100/v1.
The skill documents the critical path, the run modes, provider and template
configuration, the chaos controls, and the built-in diagnostic paths /health,
/status, and /metrics.
The skill targets vidaimock 0.3.1, and its commands were checked against that
release. It ships from this repository, so install-skills delivers it with
the other skills and no separate skill checkout is needed. See the
migration guide if an earlier deployment installed
the skill from its own repository.
The get-ai-tooling helper, which runs only when WITH_AI_TOOLING is set,
currently downloads v0.1.2, so a machine provisioned through the bootstrap runs
an older release than the skill documents and some documented commands and
flags may not be available. Install 0.3.1, for example with
cargo install vidaimock --version 0.3.1, to match.