Cascade AI Architect

Plugin cascade-ai-architect in oleksandr-rud/cascade-codex. Commands read from the repo's manifest — 6 days ago.

Skills (11)

• architect-ai-system

  Compile an unstructured AI agent or agentic-system request into a source-grounded, versioned architecture packet. Use for end-to-end agent-system design, agent topology decisions, capability-to-component decomposition, or coordinated generation of roles, skills, workflows, prompts, tools, memory, sa

• bind-agent-runtime

  Bind an accepted agent workflow to the smallest concrete host runtime, including role Context Builders, structured output, Admission, Policy Engine orchestration, persistence, recovery and optional graph execution. Use for implementation-ready runtime contracts or an audit of actual runtime ordering

• build-agent-roles

  Generate or audit reviewable role contracts for a single-agent or multi-agent architecture, with exclusive responsibilities, source and tool boundaries, outputs, handoffs, failure behavior, and evaluation assertions. Use when a selected topology needs target-harness agent definitions, when role owne

• build-agent-skills

  Map agent capability clusters into focused, triggerable, testable Codex skill design briefs and, when explicitly requested, candidate skill packages through the native skill-creator. Use when an agent architecture needs reusable skills, when an existing SKILL.md package should be redesigned or audit

• derive-persona-requirements

  Translate an existing frozen Cascade Persona agent-architecture projection into evidence-bound user-model requirements for an AI-system blueprint. Use only when architecture needs supported capabilities, contexts, constraints, failure modes, or interaction requirements; do not create or evaluate per

• design-agent-blueprint

  Turn a grounded capability map into a complete AI agent or agentic-system behavior blueprint. Use when selecting a single-agent versus multi-agent topology, defining per-agent missions and ownership, or specifying loops, state, context, memory, tools, permissions, skills, prompts, handoffs, recovery

• design-agent-workflow

  Design or audit a framework-neutral executable behavior flow for an AI agent or agentic system, including act-observe-update loops, explicit state, deterministic routing, typed handoffs, recovery, budgets, and stop rules. Use when an agent blueprint needs a workflow contract, when multi-agent orches

• map-agent-capabilities

  Extract, ground, and cluster the capabilities required by an AI agent or agentic system. Use when turning a request, brief, workflow, product spec, incident, or existing agent design into atomic outcome-oriented responsibilities; when finding missing authority or requirements; or when deciding which

• prepare-agent-evaluation

  Compile an AI-agent architecture, role, skill, workflow, or tool loop into a frozen agent-evaluation request with architecture-specific cases, assertions, profiles, rubrics, and budgets. Use before Cascade Evals execution; do not launch targets, judge outputs, reduce scores, or promote candidates he

• prepare-agent-prompt

  Compile one AI-architecture target into a source-bound prompt brief for Cascade Prompt. Use after capabilities, ownership, workflow, tools, permissions, state, recovery, stop rules, and success oracles are resolved; do not author, refine, grade, install, or promote the prompt here.

• run-improvement-cycle

  Run one bounded evidence-driven improvement cycle for a prompt, skill, agent or workflow from a measured failure and frozen baseline. Diagnose with an LLM, prepare one scoped candidate, compare matched independent evaluations and return a candidate decision; never self-promote or recursively expand

Similar plugins