Write session resume cards at end of work and restore them at session start.
93
96%
Does it follow best practices?
Impact
84%
1.23xAverage score across 3 eval scenarios
Passed
No known issues
{
"context": "Tests whether the agent reads session history from the correct canonical file path, surfaces the most recent entry (not an older one), preserves the technical detail from the card, and explicitly offers to pick up from the next step.",
"type": "weighted_checklist",
"checklist": [
{
"name": "Correct source file",
"description": "resume-output.md references `~/.claude/memory/sessions.md` as the source of the session data (not `inputs/sessions.md` or any other non-canonical path)",
"max_score": 18
},
{
"name": "Most recent entry surfaced",
"description": "resume-output.md presents or quotes the 2026-04-27 entry (debugging slow checkout / cart_service) as the session to resume — not the April 20 or April 24 entries",
"max_score": 20
},
{
"name": "Older entries not primary",
"description": "resume-output.md does NOT present the 2026-04-20 or 2026-04-24 entries as the primary session to resume",
"max_score": 12
},
{
"name": "Next step stated",
"description": "resume-output.md includes the specific next step from the most recent entry: writing/applying a database index migration and running the load test",
"max_score": 15
},
{
"name": "Offer to continue",
"description": "resume-output.md contains an explicit offer, question, or statement indicating readiness to pick up from the next step (not just a passive summary)",
"max_score": 12
},
{
"name": "Technical detail preserved",
"description": "resume-output.md includes at least one specific technical detail from the most recent card (e.g., cart_service, orders.user_id, database index, or services/cart_service.py)",
"max_score": 10
},
{
"name": "Session date referenced",
"description": "resume-output.md correctly references or uses the 2026-04-27 date for the most recent session",
"max_score": 8
},
{
"name": "No fabricated content",
"description": "resume-output.md does NOT describe a next step that was not present in the provided session history (no invented tasks or actions)",
"max_score": 5
}
]
}