Deploy an AI agent or agent harness to a Nebius Managed Kubernetes cluster using the KubeClaw production toolkit pattern (Helm chart, observability stack, secrets, autoscaling, Token Factory wiring). Produces a values.yaml, ingress config, secret-management plan, observability wiring (Prometheus + Loki + Grafana), readiness checks, autoscaling policy, and a verified rollout plan. Use whenever a user wants to deploy an agent to Kubernetes, ship an OpenClaw or NemoClaw stack to Nebius, productionize a prototype agent, asks how to host an agent in a cluster, wants a Helm chart for an LLM service, or needs autoscaling/observability for an LLM workload.
Resolve, verify, and audit an agent's identity using the AID (Agent Identity & Discovery) v2 protocol. Performs DNS-based discovery, signature verification, capability negotiation, and produces a structured identity report with trust signals. Use whenever a user wants to verify which agent is calling their service, look up an agent's capabilities and provenance from its handle, audit agent-to-agent traffic, asks 'who is this agent', 'is this AID valid', 'verify this agent signature', 'what can agent X do', or is building a service that needs to authenticate non-human callers. Trigger for agentic commerce, x402 micropayments, agent-to-agent (A2A) flows, and any system gating actions on caller identity.
Audit a business, team, role, workflow, or tool stack to identify practical AI assistant and AI agent opportunities. Use when a user wants to know what to automate, where agents could save time or increase revenue, how to prioritize AI implementation, or what first agent/skill to build.
Audit an AI agent or LLM-powered system against the agentic-engineering discipline. Scores Harness, Context, Tool Design, Skills, Memory, Evals, Model Selection, Failure Modes, and Adversarial Collaborator patterns, then prescribes three highest-ROI fixes. Use when a user wants to review an agent architecture, diagnose why an agent is unreliable or expensive, prepare a production launch, critique an LLM system design doc, or ask whether an agent is production-ready.
Comprehensive working reference for agentic engineering: the discipline of building reliable autonomous coding agents and AI-native software systems. Invoke when the user asks about designing or evaluating agent systems, harness and context engineering, tool design for agents, agent memory, evals, model selection for agentic work, the named architectural patterns (isolation, orchestration, feedback loops, failure recovery, multi-agent coordination), choosing between coding agents (Claude Code, Cursor, Devin, OpenHands, the OpenClaw ecosystem, etc.), avoiding agentic technical debt, or any question of the form 'how do I build, run, or scale agents reliably.' Use even when the user does not say 'agentic engineering' but is clearly building or operating autonomous agents.
Turn recurring tasks, SOPs, workflows, prompts, business processes, or expert know-how into reusable AI skills with trigger descriptions, step-by-step instructions, constraints, examples, and quality checks. Use when a user wants to create a Codex skill, Claude skill, agent skill, prompt pack, or reusable AI workflow.
Build a complete ambassador program — tiers, benefits, application flow, onboarding, and tracking — for developer communities, DevRel programs, or AI/web3 ecosystems. Use for ambassador program, champions program, advocate tier, referral program for developers, or builder program structure.
Audit a codebase, environment, or config for exposed API keys and secrets, and recommend safe handling, vaulting, and rotation. Scans for hardcoded keys, secrets in committed files, keys in client-side bundles, and weak storage patterns, then produces a prioritized findings report and a remediation plan. Defensive use only. Use whenever the user wants to check for leaked credentials, secure their AI/API keys, asks 'do I have any exposed keys', 'is this safe to commit', 'how should I store my API keys', or is prepping a repo to go public. Trigger before open-sourcing a repo or after suspecting a leak. Never use to locate or exploit others' credentials.