All 23 agents rebuilt: founder mindset, execution first
Every agent definition was rebuilt from scratch. Agents now execute instead of advise. Minimum viable output gates, competitive audits, and explicit done-enough criteria replace consulting theater.
What changed
This is the largest single update since the initial release. All 23 agents and 44 skills were rewritten with one mandate: ship, don't advise.
Every agent now has:
- -A founding principle (what they optimize for above all else)
- -Minimum viable output gates (what 'done' means, specifically)
- -Explicit skip/never-skip lists
- -Competitive context baked in
Highlights by agent
Helm: Infer/Ask/Decide protocol with labeled assumptions; 6-field structured handoff schema. Run /helm-brief to get a product brief that Apex can execute on directly.
Warden: Crown jewels identification before any threat modeling. /warden-threat now starts by naming what actually matters before running STRIDE.
Spine: Stripe-quality API spec output from /spine-api. Decision log required in /spine-design; no design without a rationale.
Prism: /prism-component outputs real TypeScript + Tailwind, not pseudocode. Spec-driven implementation in /prism-ui.
Cortex: /cortex-integrate runs a decision tree (prompt to RAG to tool to agent to fine-tune) before recommending an architecture. No more 'just use RAG' by default.
Apex: Reversible/irreversible action lens in /apex-plan. /apex-review gives a ship/no-ship verdict with file and line citations.
Upgrading
This is a backward-compatible update. All skill names and command signatures are unchanged. Reinstall to get the updated definitions:
claude plugin install tonone@tonone-ai