repositories / smith
smith
There are many coding harnesses - but this one is fast
owned by admin
smith-cli/tests/snapshots/e2e_help__help_eval.snap
Raw---
source: smith-cli/tests/e2e_help.rs
assertion_line: 67
expression: "String::from_utf8(help(&[\"eval\", \"--help\"])).unwrap()"
---
Run one prompt without the TUI
Usage: smith eval [OPTIONS] <PROMPT>
Arguments:
<PROMPT>
The prompt to run
Options:
--profile <PROFILE>
Exact subscription profile; required for subscription connections
--fixture-issuer <FIXTURE_ISSUER>
Explicit isolated loopback subscription fixture namespace
--credential-module <CREDENTIAL_MODULE>
Trusted credential-module package selected for this subscription only
--json
Print the structured outcome as JSON
--session <SESSION>
Explicitly resume this existing session ID
--mock
Use the scripted mock provider instead of a real endpoint
--provider <PROVIDER>
Provider kind; required for real endpoints
Possible values:
- openai: OpenAI-compatible endpoints
- openai-chatgpt-plan: `ChatGPT` plan through Responses; requires explicit profile and model
- anthropic-claude-plan: Claude subscription; explicit local profile, no API-key fallback
- google-gemini-cli-plan: Google Gemini CLI subscription; no API-key fallback
- anthropic: Anthropic Messages API
- google: Google Gemini API
- zai: Metered Z.ai API connection
- zai-coding-plan: Z.ai Coding Plan connection; never falls back to metered API billing
[env: SMITH_PROVIDER]
--model <MODEL>
Model id; required for real endpoints
[env: SMITH_MODEL]
--base-url <BASE_URL>
Endpoint base URL override
[env: SMITH_BASE_URL]
--include-usage <INCLUDE_USAGE>
Request streamed usage; explicit value replaces user and trusted project settings
[possible values: true, false]
--tool-policy <TOOL_POLICY>
Host-owned state for one qualified component tool: `name=advertised|hidden|disabled`
--tool-delegation <TOOL_DELEGATION>
Delegate a descendant directory: qualified-caller=qualified-target#directory
-h, --help
Print help (see a summary with '-h')