8 Tasks Flipped on Harness Choice Alone · DeepSeek V4 Pro (0813)
Pi Agent
DeepSeek Harness
Claude Code
OpenCode
Hermes Agent
CRM migration archive
10/10
5/10
3/10
3/10
1/2
Handover audit
11/11
6/11
6/11
10/11
6/11
Airtable formula batch
6/6
6/6
4/6
6/6
4/6
Sheets-Airtable tracker
8/10
10/10
4/10
10/10
10/10
Renewal screen
5/11
5/11
11/11
11/11
5/11
Calendar free/busy batch
10/10
10/10
10/10
10/10
2/9
Slack action items
5/5
5/5
5/5
2/5
5/5
PagerDuty timeline audit
10/10
10/10
10/10
9/10
10/10
Numbers show verifier checks cleared; a task passes only when every check passes
Source: Composio agentic eval
