Manages architecture project state in .arch/state.json and .arch/decisions.md. Activates when reading or updating project phase state, tracking component acceptance, logging decisions, or validating phase transitions.
74
85%
Does it follow best practices?
Impact
59%
1.01xAverage score across 5 eval scenarios
Passed
No known issues
{
"context": "Tests whether the agent correctly advances the project from Phase 1 to Phase 2 by writing a valid state transition to state.json, incrementing decision_count, and appending a correctly formatted entry to decisions.md.",
"type": "weighted_checklist",
"checklist": [
{
"name": "current_phase updated",
"description": "Output .arch/state.json has `current_phase` set to `\"methodology\"` (advanced from `\"evaluation\"`)",
"max_score": 25
},
{
"name": "decision_count incremented",
"description": "Output .arch/state.json has `decision_count` greater than the input value of 5",
"max_score": 15
},
{
"name": "Decision entry appended",
"description": "Output .arch/decisions.md contains a new entry that was not present in the input — verifiable by the presence of a DEC-NNN heading",
"max_score": 15
},
{
"name": "All 6 decision fields present",
"description": "The new decision entry contains all of: Decision, Rationale, Alternatives, Trade-offs, Risk, Date (as bold fields)",
"max_score": 20
},
{
"name": "Valid category used",
"description": "The new decision entry uses one of the defined categories: Requirements, Pattern, Technology, Integration, Security, Infrastructure, Process, or Reopen",
"max_score": 10
},
{
"name": "Existing decisions preserved",
"description": "Prior entries already in .arch/decisions.md are still present in the output file — none have been removed or overwritten",
"max_score": 15
}
]
}