How do I add VPP to my prompts?

Install the VPP plugin or portable Agent Skill, then start a conversation with !<tag> [--modifier ...] on line 1. The skill uses the deterministic MCP runtime to prepare, format, and validate each exchange. See the installation guide.

If the host has no skill or tool support, the custom-instructions block remains available as a reduced-assurance fallback.

What if the model ignores the footer?

The installed skill calls vpp_validate_exchange and can request structural repair before returning the result. In the custom-instructions fallback, treat a missing or malformed footer as a protocol violation and request correction.

Can I extend the tag set?

Yes, but version the extension and document new tags in your implementation notes. Keep the core tags intact so other participants can interoperate.

Does VPP work with non-OpenAI models?

Yes. Use the remote MCP endpoint when the host supports it, or the JSON/OpenAPI operations otherwise. The same deterministic core can wrap output from OpenAI, Anthropic, Gemini, or another provider.

Are tag indexes local to a locus?

No. In v1.5, each tag's index is conversation-global and continues across loci. A locus or pipeline transition resets the cycle, not the counters. Starting a new conversation with no prior state resets every counter to zero.

How should I handle multi-agent scenarios?

Carry one auditable VppState object across the orchestration boundary. Assign safe locus names for handoffs and let the runtime advance the global tag counters. Consult the normative spec when designing host-specific orchestration.