validtr run

Run task analysis, recommendation, execution, testing, scoring, and retries.

Usage

validtr run "Build a FastAPI app with JWT auth" --provider anthropic --model claude-sonnet-4-6

Flags

• --provider: anthropic, openai, gemini
• --model: required — model id (validtr has no default model)
• --compare: comma-separated providers
• --dry-run: recommendation only
• --max-attempts: retry limit
• --score-threshold: pass threshold
• --timeout: execution timeout seconds

Modes

Single provider

validtr run "Build a CLI in Go" --provider openai --model gpt-4o

Compare

validtr run "Build CRUD API" --compare anthropic,openai,gemini --model <shared-model-id>

Behavior:

• providers run sequentially
• provider with missing API key is skipped
• the single --model id is applied to every provider, so compare currently only works for a model id the providers share; per-provider model selection is a planned follow-up

Dry-run

validtr run "Automate PR code reviews" --provider anthropic --model claude-sonnet-4-6 --dry-run

Behavior:

• no execution or test containers
• returns analyzed task + recommended stack

Environment

• provider API key must be set for selected provider
• optional TAVILY_API_KEY enriches recommendation search