harness-engineering - Design harness-driven multi-agent research systems.
Design and implement harness-driven multi-agent research systems using AGENTS.md and the plan-execute-verify-reflect loop.
Tags
Updated: 2026-09-18Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Create or edit AGENTS.md files
- Design plan execute verify reflect loops
- Build MCP harness tool components
- Configure quality thresholds and failures
- Implement long transaction research patterns
Inputs
- AGENTS.md specification file
- Research query
- Codebase context
- MCP tool servers
Outputs
- Final research report with metadata
- Session state records
- Verification scores and failure traces
Requirements
- AGENTS.md compatible agent
- Model Context Protocol support
- LLM for evaluation and scoring