Agent Architect

Audit and improve an agent at the right layer.

agent-architect diagnoses where an agent problem actually belongs — persona, rules, memory, or skills — and recommends the smallest justified fix instead of a broad rewrite.

What it does

Four lanes:

1. Persona / tone — identity, voice, style
2. Rules — operating constraints, decision protocols, escalation boundaries
3. Memory — what is stored, when, how it is retrieved
4. Skills — reusable workflows, procedural packages, tool-routing

For any diagnosis, it returns:

• the weak lane
• the recommended fix type (no change / plain edit / memory tweak / rule change / skill change)
• where the patch belongs
• what not to touch

Why this exists

Most agent improvement efforts fail because they change the wrong layer:

• editing tone when the real problem is a missing rule
• adding a skill when the agent just forgot a standing fact
• rewriting memory when the workflow itself is missing

agent-architect routes the fix to the right place first.

Installation

ClawHub

clawhub install agent-architect

Manual

Copy the agent-architect/ folder into your skills directory.

Works well with

• skill-builder — when the result is "tighten or add a skill"
• battle-tested-agent — when the diagnosis suggests reliability hardening
• cognition — when the issue is memory architecture
• openclaw-guide — when the issue is platform behavior, not the agent itself

License

MIT