agent-plugin-architectlisted

# Agent Plugin Architect Design or audit the smallest coherent packaged-plugin control plane for Codex and Claude Code. Treat Codex as the first-class target and preserve a single shared Skill tree when both hosts can discover it without weakening Codex. ## Route The Work 1. Confirm that the request explicitly concerns packaged agent-plugin architecture. Repo-local `AGENTS.md` or `.agents/skills` work belongs to `agents-architect`; ordinary plugin code or documentation has no Axiom route solely because it lives in a plugin repository. 2. Inventory the package before editing. Read `references/package-inventory.md`. 3. Define route ownership and trigger boundaries with `references/route-and-trigger-contracts.md`. 4. For shared Skill structure and direct discovery, read `references/packaged-skill-architecture.md`. 5. For startup hooks, untrusted content, credentials, or runtime effects, read `references/hooks-and-trust-boundaries.md`. 6. For manifests, wrappers, and Codex/Claude Code parity, read `references/cross-host-packaging.md`. 7. For routing cases, context cost, or host observations, read `references/evaluation-and-evidence.md`. 8. Before completion, read `references/validation-reporting.md`. Load only references needed for the active phase. A request may compose this route with at most one other Axiom route when both contracts are independently necessary. An explicit usage-cost goal may add `optimize-codex-usage`; retrospective review may a