[plan] Phase 5: Add runtime config-renderer hook to CodingAgentEngine interface

[github-actions]

Objective

Add a config-renderer hook to the runtime adapter interface so runtimes can emit generated config files and metadata. This allows engines like OpenCode (which use provider/model config files rather than just flags and env vars) to participate in the architecture cleanly.

Context

Issue #20416 Phase 5: some runtimes need to write config files to disk before execution (e.g., OpenCode's ~/.opencode/config.toml). Currently, all runtime configuration happens through env vars and command flags. This hook decouples config generation from execution step generation.

Prerequisite: Phases 1–4 should be merged first, though this phase is less tightly coupled than the others.

Approach

1. Add RenderConfig to the CodingAgentEngine interface in pkg/workflow/agentic_engine.go

// RenderConfig optionally generates runtime config files or metadata.
// Returns a slice of GitHub Actions steps that write config to disk.
// Implementations that don't need config files should return nil, nil.
RenderConfig(target *ResolvedEngineTarget) ([]map[string]any, error)

2. Implement no-op RenderConfig in all existing runtime adapters

In pkg/workflow/claude_engine.go, codex_engine.go, copilot_engine.go, gemini_engine.go:

func (e *ClaudeEngine) RenderConfig(target *ResolvedEngineTarget) ([]map[string]any, error) {
    return nil, nil
}

This preserves backward compatibility — no behavior change for existing engines.

3. Call RenderConfig from the orchestrator

In pkg/workflow/compiler_orchestrator_engine.go, after resolving the engine target, call runtime.RenderConfig(target) and prepend any returned steps before the execution steps.

4. Add a test runtime that uses RenderConfig

In test code only, create a minimal configRenderingEngine that writes a dummy config file step. Test that the orchestrator includes those steps in the correct position (before execution steps).

5. (Optional) Add a skeleton OpenCode engine

If time permits, add a skeleton pkg/workflow/opencode_engine.go that:

• Registers in the catalog as opencode
• Implements RenderConfig to emit a step that writes an OpenCode config file
• Does not need to be feature-complete — just validates the hook works end-to-end

This is optional and can be a separate follow-up issue if it grows too large.

Files to Modify

• Modify: pkg/workflow/agentic_engine.go — add RenderConfig to interface
• Modify: pkg/workflow/claude_engine.go, codex_engine.go, copilot_engine.go, gemini_engine.go — add no-op implementations
• Modify: pkg/workflow/compiler_orchestrator_engine.go — call RenderConfig
• Create: pkg/workflow/engine_config_render_test.go
• Optionally create: pkg/workflow/opencode_engine.go
• Run: make agent-finish

Acceptance Criteria

• RenderConfig method added to CodingAgentEngine interface
• All four existing engines implement RenderConfig returning nil, nil
• Orchestrator calls RenderConfig and prepends returned steps
• Test runtime with non-nil RenderConfig output produces correct step ordering
• All existing workflow compilation outputs unchanged (no regression in .lock.yml files — run make recompile and verify no diff)
• make test-unit passes
  Related to GitHub Agentic Workflow Engine Enhancement Proposal #20416

Generated by Plan Command for issue #20416 · ◷

• expires on Mar 13, 2026, 2:45 AM UTC