Viable Prompt Protocol (VPP)

Appearance

Why VPP

VPP is for transcripts that must stay readable, auditable, and replayable as projects get longer, ideas get denser, and decisions become costly.

A shared contract for how humans and assistants structure multi-turn work—on any model you already use.

Try VPP in your system prompt Read the spec
VPP is not a chatbot in disguise. It does not swap models or pretend one system is another. VPP is a protocol layered on top of vanilla LLMs: a shared contract for how humans and assistants structure multi-turn work.

What breaks in normal chat ​

  • Intent drifts: the “mode” of a turn (explore, draft, critique) stays implicit.
  • Corrections get lost: fixes are made, but the transcript stops being traceable.
  • Assumptions hide: choices accumulate without being named or counted.
  • Review becomes guesswork: collaborators can’t reliably replay reasoning.

What VPP adds ​

  • Explicit intent on every user turn via !<tag>.
  • Mirrored intent on every assistant turn via <tag>.
  • Auditability via a footer that records cycle, sources, and assumptions.
  • Iterative control via a loop that makes correction and refinement legible.

Where VPP shines ​

  • Long, big projects: the transcript stays navigable months later.
  • Complex ideas: you can separate grounding, probing, drafting, and critique.
  • Recursive prompting: when you don’t yet know what you want, the loop supports discovering it together—without losing structure.

A micro-example ​

text
!<q>
What are the unknowns in this plan?

<q>
Unknowns:
- What “success” means (deliverable + audience).
- Which constraints are hard vs negotiable.
- What data we can cite vs must infer.
[Version=v1.4 | Tag=<q_1> | Sources=<none> | Assumptions=0 | Cycle=1/3 |
Locus=unknowns]

!<c>
Clarify the deliverable and remove the weakest assumption.

<c>
Understood. Deliverable: a single-page spec with a join-first CTA.
Removed assumption: that we can infer the user’s audience without
asking.
[Version=v1.4 | Tag=<c_1> | Sources=<none> | Assumptions=0 | Cycle=2/3 |
Locus=spec-tighten]

The tag states intent. The footer makes progress reviewable.

Why prompt templates aren’t enough ​

Prompt templates are great at starting a session. They do not enforce structure across dozens of turns.

VPP makes structure continuous: Every turn declares intent, every response mirrors it, and the transcript stays auditable as it grows.

Try VPP anywhere

Paste the header snippet into your system prompt, then begin with !<g> or !<q>.

Next ​

Last updated: