Research: Autonomous Multi-Agent Orchestrator
Status: Complete (leverages feature 019 research)
Summary
This feature builds directly on the comprehensive research conducted in feature 019 (Autonomous Multi-Agent Orchestration Research). No additional research was required.
Key Findings from Feature 019
Agent CLI Capabilities
- 9 of 12 agents have CLI support suitable for autonomous orchestration
- 8 Tier-1 agents (ready without workarounds): Claude Code, GitHub Codex, GitHub Copilot, Google Gemini, Qwen Code, OpenCode, Kilocode, Augment Code
- 1 Tier-2 agent (requires workaround): Cursor (needs timeout wrapper)
- 3 agents not suitable: Windsurf (GUI-only), Roo Code (no official CLI), Amazon Q (transitioning)
Feasibility Verdict
Fully Feasible - All quality gates passed:
- QG-001: ≥6 agents with CLI (9 found)
- QG-002: Cursor CLI documented with workaround
- QG-003: All findings include source links
- QG-004: Parallel constraints documented
Architecture Recommendation
The research recommended a Python-based orchestrator with:
- Scheduler: Reads WP dependencies, assigns agents
- Executor: Spawns agent processes via asyncio
- Monitor: Detects completion via exit codes and JSON
- State Manager: Persists state for resume
Minimum Viable Agent Set
For initial implementation:
- Implementation: Claude Code (best task input support)
- Review: GitHub Codex (different perspective)
- Fallback: OpenCode (multi-provider flexibility)
References
- Feature 019 Spec
- Feature 019 Data Model
- Feature 019 Research Findings
- Sample agents.yaml
Technical Decisions
All technical decisions for this feature are derived from the 019 research:
| Decision | Source | Rationale |
|---|---|---|
| Python + asyncio | 019 Architecture | Native async, subprocess support |
| JSON state file | 019 Data Model | Human-readable, git-friendly |
| Priority-based selection | 019 Config Schema | User control, fallback support |
| Per-agent invokers | 019 CLI Matrix | Different flags per agent |