Vericore

PR Intelligence

PR Intelligence converts a local Git change set into a deterministic engineering assessment. It combines Git diff metadata, dependency impact, existing engineering-risk signals, package boundaries, architecture signals, and likely test candidates.

CLI

The complete command syntax is maintained in CLI Reference. The PR Intelligence-specific forms are:

vericore pr-intelligence /path/to/repository --json

vericore pr-intelligence /path/to/repository \
  --base main \
  --head feature/my-change \
  --json

The JSON report is written to output/pr-intelligence.json when JSON output is requested.

--base and --head must be supplied together. If neither is supplied, the current working tree is analyzed.

REST

POST /pr-intelligence accepts a local repository path and optionally a base/head revision pair. Remote repository URLs are deliberately rejected by the current local endpoint.

{
  "repoPath": "/workspace/example",
  "baseRevision": "main",
  "headRevision": "feature/my-change"
}

Omit both revisions to analyze the working tree. Supplying only one revision is rejected.

Result

The result is versioned and machine-readable:

Every finding has a stable rule identifier, severity, affected repository-relative paths, evidence values, and a human-readable reason.

Rules

Rule Meaning
CHANGE_UNRESOLVED A changed/deleted path could not be resolved in the analyzed graph.
IMPACT_BROAD The dependency blast radius is substantial.
ARCH_CROSS_PACKAGE Impact crosses the package boundary of changed files.
CHANGED_HIGH_RISK_COMPONENT A high/critical deterministic risk component was modified.
TEST_CANDIDATE_MISSING No likely test candidate was identified.
CHANGE_LARGE The change is large by file or line count.

These are review signals. They do not prove runtime correctness, test sufficiency, or production impact.

Evidence-first workflow

Git working tree / revision pair
              ↓
       GitChangeSetBuilder
              ↓
       PRIntelligenceAnalyzer
              ↓
       Impact / Risk / Architecture / Test signals
              ↓
        Versioned PR result
              ↓
      Grounded evidence / planner / AI

The deterministic core has no dependency on an LLM or GitHub API. The CLI and server can therefore use the result as a stable review signal even when AI is disabled. This makes it suitable for CI and allows AI layers to explain evidence without becoming the source of truth.

Current boundary

PR Intelligence currently produces deterministic review signals and machine-readable evidence. It does not claim to prove runtime correctness, test coverage, deployment safety, or production impact, and it does not autonomously approve, merge, or modify a change.