the suite
Six tasks
One task per category, each asking a different question about what a coding agent can actually do. Every task page below has the verbatim prompt, the fixture it starts from, and every score it has produced across releases.
Backend
Inventory service
Can it build a service to a contract?
Leader: claude · haiku · low 77.0
Bug Fix
Fix date range overlap bug
Can it fix a reported bug without breaking anything?
Leader: claude · haiku · low 100.0
Frontend
Issue board filters
Can it extend an existing app in its own conventions?
Leader: codex · gpt-5.6-sol · low 68.3
Planning Audit
Codebase audit, planted defects
Can it read a codebase critically and plan?
Leader: claude · haiku · low 79.0
Signature
shipshit.dev landing hero
The same visual brief every release.
Leader: codex · gpt-5.6-sol · low 92.5
Taste
Landing page, five themes
Left alone, the agent invents a small project (SaaS, developer tool, one-product store, newsletter, service), builds five genuinely different landing pages for it that follow conversion practice, and recommends one. Tests design judgment, divergence, copy and self-assessment rather than execution of a given brief.
Leader: codex · gpt-5.6-sol · low 100.0
UX/UI
Shipcut pricing page
Can it design, not just code?
Leader: codex · gpt-5.6-sol · low 100.0