What it does

Auditors don't need another list of findings. They need to know how a finding is reached: which entry-point functions can call into a vulnerable function, what state that function touches, and what dangerous sink sits at the end of the path. AuditGraph connects Slither's facts into a graph and answers path questions directly: blast radius (everything within 4 hops of a function), paths to a sink (every entry path to call.value, delegatecall, selfdestruct, tx.origin, or an unprotected initializer), who can reach a privileged function (all callers, transitively), and findings by severity and CWE. The demo ships two contracts with the same shape: Vault (deliberately vulnerable) and VaultSafe (the guarded negative control). Slither finds a High reentrancy finding in Vault and none in VaultSafe; the graph shows why.

How we built it / how it uses Relate

Mode: single file (data/audit.rl). Why: the whole audit graph is one file you can email, check into git, or hand to a judge, who can open it with relate query, serve it with relate serve, or load it in a browser via the WASM build. No server to install, no signup, no shared database — the audit travels with the repo it describes. In-memory mode would vanish when the process exits, so judges could not inspect it. Ingest is Slither JSON + Slither Python API → Cypher → Relate (relate-db). The viewer is a static page (no build step) that talks to the local relate serve HTTP Query API v2 and renders results as a table plus an SVG path diagram. Verified outputs: 39 nodes, 47 relationships; paths-to-sink returns drain → withdraw → call.value and emergencyWithdraw → withdraw → call.value; 14 tests passing; HTTP API returns 200 with auth, 401 without.

Honest limits

One writer per store; the graph must fit in RAM; no auth on the store file itself; findings are only as good as Slither; unprotected-init is a documented heuristic (Rule U1), not a Slither detector.

Built With

Share this project:

Updates

Submission history