agent-creationlisted

# Agent Creation ## What is an agent? A specialist agent is a contract over project-context plus a small identity that points at a domain. The contract is the load-bearing part: a `read` list that filters which slices of project-context get injected (the token-efficiency lever) and a `write` list that confines what the agent may persist back (the security lever). The identity that wraps the contract is mostly *shared* — for the builder agents (`developer`, `platform-architect`, `gitops-operator`) it is nearly the same essence — and the only genuinely per-agent pieces are the contract, the skills, and a small subset of "what this agent builds." If the component you are building has no distinct contract, no delegation surface, and could work as injected text, it is a skill, not an agent. That decision belongs upstream — this skill assumes it has been made. ## The contract-first model Creating an agent is **contract + skills + a small domain subset**, not a personality authored from scratch. Three facts drive every decision below: 1. **The contract is the first design decision.** Before identity, before tools, before skills, decide the `project_context_contracts` block. The `read` list is what makes the agent token-efficient — it filters the project-context injection down to the slices this domain actually reasons over, so the agent is not paying for context it never reads. The `write` list is what makes the agent safe — it is the allowlist the runtime checks before accepti