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 saves the session resume card to the correct global location, uses the correct date-prefixed format with em-dash and colon separators, writes a single-line specific entry, and avoids using memory.md for session data.",
"type": "weighted_checklist",
"checklist": [
{
"name": "Correct file path",
"description": "session-confirmation.md references `~/.claude/memory/sessions.md` as the save location (not a relative path, repo path, or any other location)",
"max_score": 18
},
{
"name": "Not repo-local",
"description": "session-confirmation.md does NOT reference a local/relative path like `./sessions.md`, `sessions.md`, or a path starting with `/workspace` or the current directory as the save target",
"max_score": 8
},
{
"name": "Date prefix format",
"description": "The resume card in session-confirmation.md starts with `- YYYY-MM-DD` (a dash followed by a date in four-digit year, two-digit month, two-digit day format)",
"max_score": 12
},
{
"name": "Em-dash separator",
"description": "The resume card uses an em-dash (`—`) to separate the date from the task description (not a hyphen or colon)",
"max_score": 8
},
{
"name": "Colon separator",
"description": "The resume card contains a colon (`:`) separating the task description from the next-step detail",
"max_score": 8
},
{
"name": "Specific file or location",
"description": "The resume card mentions at least one specific file path, function name, class name, or line number relating to the debugging context (e.g. product_controller.rb, includes(:reviews), or spec file)",
"max_score": 12
},
{
"name": "Concrete next step",
"description": "The next step in the card is a specific action (e.g., apply eager loading fix, run controller test) rather than a vague statement like 'need to continue', 'will continue', or 'pick up tomorrow'",
"max_score": 12
},
{
"name": "Single-line entry",
"description": "The resume card is written as a single line (or single dash-prefixed entry), not as multiple lines, paragraphs, or a structured block",
"max_score": 8
},
{
"name": "Domain-specific description",
"description": "The card's task description references the specific technical context (N+1 query, product catalog, reviews, or API performance) rather than generic language like 'working on code' or 'fixing a bug'",
"max_score": 9
},
{
"name": "memory.md not used",
"description": "session-confirmation.md does NOT indicate that any session entry was written to `memory.md`",
"max_score": 5
}
]
}