Summary

Add real E2E smoke tests that execute Archon workflows in GitHub Actions CI, verifying that the full workflow engine works end-to-end with both Claude and Codex providers.

Motivation

Current CI only runs unit/integration tests with mocks and a Docker health check. There's no verification that actual workflow execution works — provider selection, structured output, tool use, skills, MCP, script nodes, etc. A regression in any of these would only be caught manually.

Scope

New GitHub Actions workflow (.github/workflows/e2e-smoke.yml)

• Triggers on push to main/dev only (no PR trigger to avoid API cost abuse)
• 4 parallel jobs: deterministic, claude, codex, mixed-providers
• Installs Claude CLI, Codex CLI, uv for Python scripts

E2E workflow files

• e2e-deterministic.yaml — bash, script (bun), script (uv), conditions, trigger rules (0 API calls)
• e2e-claude-smoke.yaml — enhanced with command node, model overrides
• e2e-codex-smoke.yaml — with explicit model config
• e2e-mixed-providers.yaml — cross-provider output refs
• e2e-all-nodes.yaml — fix broken script node
• e2e-skills-mcp.yaml — skills injection, MCP server, effort, systemPrompt

Supporting files

• Test skill, MCP config, test command, Python script

Cost

• ~$0.01-0.02 per run using haiku (Claude) and gpt-5.1-codex-mini (Codex)

Requirements

• ANTHROPIC_API_KEY and OPENAI_API_KEY GitHub secrets must be configured