Standing Expert Agent Fleet
Multi-Agent Architecture
Designed the /ascend forging protocol, built the dispatch routing layer, and forged all standing agents from first principles in their respective domains.
Project Overview
The standing agent fleet is the capability layer of the Chiron harness. Each agent is forged through /ascend: a structured process that builds a glossary node, codex knowledge graph, grounding artifacts, and a final SKILL.md before the agent is deployed. Agents cover SEO/AEO/GEO, MES futures trading, cold outreach, Convex database engineering, frontend motion, AI-visibility synergy, agentic control-plane architecture, and more. The fleet is routed via a dispatch skill that classifies the task type and selects the most grounded agent, keeping generalist fallback to a minimum.
Challenge
Build a multi-agent fleet that compounds in capability over time without accumulating duplicated scope, stale expertise, or untested agents running in production domains.
Key Features
/ascend Forging Protocol
Each agent earns its domain through /ascend: glossary -> codex node -> grounding artifact -> SKILL.md -> deployment. No ungrounded domain claims.
Skill-First Dispatch
A routing layer classifies every task and dispatches to the most grounded specialist; generalist fallback is the last resort, not the default.
Composable, Non-Duplicating
Agents are scoped by domain and composed via explicit handoffs; each owns its boundary and delegates cross-domain work rather than trying to cover it.
Lessons Learned
- school A forged agent with a codex node outperforms a prompted generalist on every repeat task. The investment in /ascend pays back within 3 arcs.