clauz3-authoringDesign or extend a Clauz3 trusted library when agents need capabilities or safety contracts that existing libraries do not provide. Define trusted effects, typed results, preconditions, contracts, shared events, runtime obligations, tests, and installable metadata. Use for library authoring, not for
clauz3-exploringUse when the user asks for hands-on, open-ended testing of clauz3 by driving an independent Claude Code agent through real tasks — picking a target repo, issuing prompts, watching what the agent submits, deciding approvals as the user-proxy, escalating to adversarial cases, and reporting findings. T
clauz3Use when an agent needs to act in the real world — send mail, edit a file, fetch a URL, move money, run a lab step, modify any external system — and finds that no direct tool exists for it. If the project ships a clauz3 trusted layer, that IS the tool, but reached only by writing a small Python prog