I realized my AI agent (Clawdbot) had a flaw: it was too eager to please. When I asked for a tool, it would jump straight to coding the simplest version, often skipping design, architecture, and robustness.

The Anti-Pattern

I asked for a Spec Manager skill. The agent immediately wrote a Python script that just took a title and status. It missed the entire point of a spec workflow—capturing requirements, parsing ambiguity, and creating detailed plans.

It optimized for Done instead of Quality. It lacked Governance.

The Soul Upgrade

We decided to fix this at the system level. We modified the agent identity files (SOUL.md and AGENTS.md) to install mandatory protocols.

  • Superpowers Protocol: Before writing code, MUST invoke brainstorming and writing-plans.
  • Architecture Protocol: Before building tools, MUST consult agent-native-architecture principles (Parity, Granularity, Emergence).
  • Knowledge Protocol: When a problem is solved, MUST invoke compound-docs to capture the solution.

Why This Matters

This shifts the dynamic from Master/Servant to Partners. By encoding governance into the agent soul, we ensure every future task benefits from system thinking.