Skip to content

feat: dbt PR review — layered (deterministic + advisory) reviewer for dbt/SQL PRs #855

@anandgupta42

Description

@anandgupta42

Summary

Ship a AI code reviewer specialized for dbt/SQL, producing a single signed verdict (APPROVE / COMMENT / REQUEST_CHANGES) where every blocking finding is backed by a deterministic engine call, not a model opinion.

Motivation

Generic AI reviewers miss dbt/SQL-specific risk: query non-equivalence in "refactors", column-lineage/DAG blast radius, PII exposure, warehouse-cost anti-patterns, and dbt config/contract changes. We have the deterministic engine (altimate-core) to prove these — wire it into a PR reviewer.

Scope

  • Layer 1 (authoritative, the only layer that blocks): altimate-core over parsed SQL ASTs — query equivalence, column lineage / blast radius, PII classification, A–F grade, AST anti-pattern lint.
  • Layer 2 (fallback): thin self-verifying catalog for raw-diff/Jinja signals not yet AST.
  • Layer 3 (advisory, never blocks): an LLM reviewer grounded in layers 1–2 for intent/naming/business-logic/test-coverage commentary, clamped to ≤ warning.
  • Signed, replayable verdict envelope keyed to the dbt manifest; CLI + GitHub Action; comment and gate modes.

Acceptance

  • Deterministic verdict + advisory comments on a PR; only layer-1 engine findings can block.
  • Validated against a real dbt repo's PR history (catches a real non-equivalence bug, FP-safe on intentional refactors).
  • Depends on @altimateai/altimate-core >= 0.4.0.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions