arch-playground/AI Workflow Runner

Run AI workflows with Node.js, Python, and Java runtime support

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
workflow_pathPath to the workflow.md file (relative to workspace root). Required unless list_models is true.no""
promptInput prompt to pass to the workflow (max 100KB)no""
env_varsJSON object of environment variables (max 64KB, 100 entries)no{}
timeout_minutesMaximum execution time in minutes (default: 30)no30
validation_scriptOptional validation script (file path like check.py or inline with prefix like python:print("ok")). Receives AI_LAST_MESSAGE env. Return empty/true for success, other string to continue workflow.no""
validation_script_typeScript type for inline scripts: python or javascript. Auto-detected for file paths by extension (.py or .js). Only required when validation_script is set and type cannot be auto-detected.no""
validation_max_retryMaximum validation retry attempts when script returns non-empty/non-true (default: 5, max: 20)no5
opencode_configPath to OpenCode config.json file (relative to workspace, or absolute under runner.temp/tmp). Contains provider and model settings.no""
auth_configPath to OpenCode auth.json file (relative to workspace, or absolute under runner.temp/tmp). Contains API keys and authentication. Store in GitHub Secrets, not Variables.no""
modelModel to use for AI execution (e.g., "anthropic/claude-3-opus"). Overrides config file default.no""
list_modelsIf "true", print available models and exit without running workflownofalse
debug_logEnable verbose debug logging to file. Also activated by ACTIONS_STEP_DEBUG=true or RUNNER_DEBUG=1.nofalse
debug_log_pathPath for debug log file. Defaults to $RUNNER_TEMP/opencode-debug.log. Accepts workspace-relative or absolute paths under RUNNER_TEMP/tmp.no""
namedescription
statusExecution status: success, failure, cancelled, or timeout
resultWorkflow execution result as JSON string (max 900KB)