What VibePro Is

VibePro supports investigating what should change, then carrying that reasoning into implementation and PR review. It connects the user need, decision evidence, agreed behavior, implementation references, and check results.

AI coding agents can produce technically valid code while solving the wrong problem. VibePro addresses that product-intent gap with a small set of repository-local records:

Request and expected outcome
  -> When needed: frame questions <-> investigate <-> reconsider with evidence
  -> Adopt a direction under human and repository authority
  -> Story (need and acceptance criteria)
    -> Spec (behavior, invariants, code/test references)
      -> Implementation
        -> Verification and review records
          -> PR summary

This is a conceptual connection, not a mandatory command order. Investigation can start from an existing Story, and its results do not automatically finalize a Story or Spec.

What each record contributes

• Investigation and judgment retain questions, evidence, options, and unresolved issues. The AI host and judgment investigate support investigation and reconsideration; judgment prepare --investigation carries candidates into an input draft, not automatic adoption. See Senior Engineering Judgment.

• Story states the user need and the acceptance criteria that make the change worth doing.

• Spec turns that need into concrete behavior and invariants, with code_refs and test_refs pointing toward implementation and verification.

• Verification records what was run or recorded against the repository state. It is evidence about a check, not a promise that the product is correct.

• Review records the inspection performed by a person or reviewer role, including the inspected inputs and the result.

• PR preparation projects the recorded Story, Spec, verification, review, trace, and decision context into a machine-readable summary and PR body.

The records are deliberately close to the repository. A later reviewer can follow the reason for the change without relying on an agent's private conversation or an unrecorded handoff.

What the current core does

The current beta can:

• initialize a repository-local .vibepro/ workspace and choose an output language;
• add, select, and diagnose Stories;
• prepare investigation requests for the AI host, read explicitly supplied existing Graphify artifacts and source, and revisit judgments with new evidence;
• write draft or final Specs and check their structural references;
• declare or inspect traces and explicit decisions when the change needs them;
• run or record unit, integration, end-to-end, typecheck, or build verification;
• prepare and record a lightweight review; and
• prepare a PR summary and body for an ordinary pull-request workflow.

The installed binary's vibepro help output is the authoritative command contract. See Install and First Run for a concrete sequence.

Draft Specs can be used while shaping a change. A final Spec has stronger readiness preconditions: Graphify context, a Story diagnosis, and readiness evidence against the current HEAD. See the control loop for the order of those checks.

What is checked, and what is not

VibePro can check structural facts such as whether a declared file, symbol, or test reference exists, whether a reference has the expected anchor, and what a recorded verification command returned. Those checks make evidence easier to inspect.

The AI host also interprets evidence and generates questions and options. judgment investigate returns a model_request and validates responses, but does not autonomously call models, generate Graphify artifacts, or fetch through external connectors. Generate Graphify separately with the explicit vibepro graph . --run-graphify command, or pass an existing artifact to the investigation. New evidence requires reinterpretation rather than reuse of the old recommendation. Missing evidence remains unknown, partial, or unavailable.

This does not establish semantic correctness. Graphify edges are structural candidates, not proof of contracts or runtime success. A code reference may point to the wrong behavior, and a passing test may cover the wrong scenario. Human product and engineering review and adoption responsibility remain necessary.

Likewise, VibePro does not implement application code, certify safety, approve a pull request, or merge code. Repository policy, CI, and people retain those authorities.

Cost and fit

Story and Spec records are an explicit maintenance cost. Keep them when the connection between a product need and a code change is valuable; update them when the intended behavior changes. They complement README files, issues, design documents, and normal code review—they are not a claim that those forms of context can be replaced by one generated artifact.

The minimal core also does not include the former broad Gate DAG, managed execution controller, review-lifecycle accounting, delivery-efficiency budgets, or automatic audit bundles. The current judgment DAG provides optional investigation and reconsideration support. Unlike the retired gates, it does not automatically certify intent alignment or hold adoption, execution, or merge authority.

Authority and history

The public manual describes the current beta, while the version history may mention older releases. Treat the installed vibepro help output as the command authority and package.json as the version authority.