Self-Enforcing
Workflows

AI agent skills that verify their own work.

A design pattern where every skill carries its own Definition of Done and calls a universal quality gate as its final step. No external orchestration required.

AI agents are optimistic completers.

They declare success when they are 80% done. A skeleton-project skill would scaffold a folder, push to GitLab, and declare done. Half the time the remote push had silently failed. The folder existed. The commit existed. The remote did not.

Self-Enforcing Workflows are the fix.

External enforcement

"Someone will check your work before it ships."
Skill runs Declares done Supervisor checks (maybe)
vs

Self-enforcement

"You check your own work before you hand it off."
Skill runs Calls auditor Verified ✓