Enforces a 4-phase architecture design workflow by reading `.arch/state.json` on every request to gate responses by phase. Phase 1 extracts and validates requirements from PRDs; Phase 2 selects architecture patterns and establishes high-level structure; Phase 3 designs and accepts components sequentially; Phase 4 finalises and documents the solution. Use when discussing system design, solution architecture, PRD analysis, component design, technology selection, or architecture patterns — distinct from general coding help by its strict phase-gating, anti-pattern detection, and state-tracked component acceptance.
93
97%
Does it follow best practices?
Impact
89%
1.07xAverage score across 5 eval scenarios
Passed
No known issues
{
"context": "Tests whether the agent correctly enforces Phase 1 boundaries by refusing to engage with technology and architecture pattern questions, redirecting to completing requirements gathering instead, while remaining constructive about what comes next.",
"type": "weighted_checklist",
"checklist": [
{
"name": "No tech recommendation",
"description": "response.md does NOT endorse or recommend any specific technology named in the developer's message (e.g. Kafka, PostgreSQL, React) or any alternative technology",
"max_score": 20
},
{
"name": "No architecture pattern",
"description": "response.md does NOT recommend or evaluate microservices or any other architecture pattern",
"max_score": 15
},
{
"name": "No component design",
"description": "response.md does NOT describe component structures, service boundaries, or implementation details",
"max_score": 15
},
{
"name": "Explicit redirect",
"description": "response.md explicitly tells the developer that technology decisions cannot be made yet and directs them back to completing Phase 1 requirements first",
"max_score": 20
},
{
"name": "Phase gate referenced",
"description": "response.md references the current phase state, incomplete requirements, or the phase gate as the reason for deferring tech decisions",
"max_score": 15
},
{
"name": "Future path stated",
"description": "response.md tells the developer what will happen after Phase 1 is complete (i.e. that technology decisions will be made then)",
"max_score": 15
}
]
}