@@ -94,9 +94,13 @@ export async function runGitEvals2(
9494 }
9595
9696 // Save trace to logs directory
97+ const safeSpec = commit . spec
98+ . split ( '\n' ) [ 0 ]
99+ . replace ( / [ ^ a - z A - Z 0 - 9 ] / g, '_' )
100+ . slice ( 0 , 30 )
97101 const safeAgentId = agentId . replace ( / [ ^ a - z A - Z 0 - 9 - ] / g, '_' )
98102 const safeCommitShort = commit . sha . slice ( 0 , 7 )
99- const traceFilename = `${ safeAgentId } -${ safeCommitShort } .json`
103+ const traceFilename = `${ safeSpec } - ${ safeAgentId } -${ safeCommitShort } .json`
100104 const tracePath = path . join ( logsDir , traceFilename )
101105
102106 const traceData = {
@@ -206,13 +210,11 @@ export async function runGitEvals2(
206210 console . log ( '\n=== Summary ===' )
207211 for ( const [ agentId , data ] of results ) {
208212 console . log ( `\n${ agentId } :` )
209- console . log ( ` Average Score: ${ data . averageScore . toFixed ( 2 ) } /10` )
210- console . log ( ` Average Cost: $${ data . averageCost . toFixed ( 4 ) } ` )
213+ console . log ( ` Score: ${ data . averageScore . toFixed ( 2 ) } /10` )
214+ console . log ( ` Cost: $${ data . averageCost . toFixed ( 4 ) } ` )
215+ console . log ( ` Duration: ${ ( data . averageDuration / 1000 ) . toFixed ( 1 ) } s` )
211216 console . log (
212- ` Average Duration: ${ ( data . averageDuration / 1000 ) . toFixed ( 1 ) } s` ,
213- )
214- console . log (
215- ` Success Rate: ${ data . runs . filter ( ( r ) => ! r . error ) . length } /${ data . runs . length } ` ,
217+ ` Success: ${ data . runs . filter ( ( r ) => ! r . error ) . length } /${ data . runs . length } ` ,
216218 )
217219 }
218220
0 commit comments