How it works
Multi-model analysis with architecture context
Guardix is not a single prompt against your code. It's a pipeline that understands your system architecture, runs checklist-driven analysis across multiple models, and validates findings through consensus.
The pipeline
Five stages from repository to report.
Clone & scope
Guardix clones the exact branch and commit from GitHub. Only in-scope Solidity files enter the pipeline.
Architecture understanding
The system maps contracts, inheritance, call graphs, and state flows. It extracts invariants, assumptions, and design decisions as first-class artifacts.
Multi-model analysis
Multiple independent model perspectives analyze the codebase in parallel — each runs checklist-driven analysis and produces findings without seeing the others.
Consensus validation
Findings are cross-validated across models. Only issues with multi-model agreement and high confidence reach the final report.
Report generation
Validated findings, architecture artifacts, and severity assessments are assembled into a versioned, shareable report.
Stage 2
Architecture understanding — not just pattern matching.
Before looking for issues, the pipeline maps contracts, inheritance graphs, call flows, and state transitions. It extracts invariants, assumptions, and design decisions as first-class artifacts that appear alongside findings in the report.
Total supply must equal sum of individual balances at all times
Threat if broken
Direct loss of depositor funds
Queue state transitions must be monotonic: pending → processing → done
Threat if broken
Double-processing of withdrawals
Oracle price feeds assumed fresh within 1-hour staleness window
Threat if broken
Incorrect liquidation at stale prices
Admin role changes cannot bypass timelock constraints
Threat if broken
Unauthorized privilege escalation
Unchecked external callback before balance update
Admin rotation bypasses role boundary
Invariant drift between queue state and accounting
Unchecked transfer return value hides failure
Stages 3–4
Multi-model analysis + consensus validation.
Multiple independent model perspectives analyze the codebase in parallel using checklist-driven methods. Findings are then cross-validated — only issues with multi-model agreement and high confidence reach the final report.
Broad
Parallel analysis
Multi
Model perspectives
Consensus
Validation
Stage 5
Versioned reports you can share and act on.
Each scan produces a versioned report tied to the exact commit. Share a link or export a PDF. After fixing issues, re-audit on a new commit and compare scan versions side by side.
stonevault/vault-core
share linkShare a versioned report that stays tied to the scan version — stakeholders see the same data engineers are reviewing.
See the pipeline in action.
Start a free audit to experience the full pipeline from repository to report.