The trust layer for AI-written code. AI writes, SPARDA proves.
Bring SPARDA's verdict into your editor. Deterministic, offline behavior proofs — no API key, no
cloud, nothing leaves your machine. The extension is a thin wrapper over the sparda-mcp CLI, so
its verdict never drifts from the CLI's.
What it does
SPARDA: Prove workspace — compiles your backend and shows the trust verdict
(PROVEN / PARTIAL / …) in the status bar and a notification.
SPARDA: Apocalypse — surfaces every finding (unguarded mutation, BOLA/IDOR, irreversible
effect, …) in the Problems panel at its exact file:line.
SPARDA: Gate — arm the baseline — freezes the current behavior as the accepted baseline.
SPARDA: Install the gate into Claude Code — wires sparda gate into Claude Code's
PostToolUse hook, so every AI edit is proven in the loop.
Requirements
The sparda-mcp CLI must be reachable. By default the extension runs npx --no-install sparda-mcp;
point it at a global install or a custom path via the sparda.command setting.
Privacy
100% local. The extension only ever runs the SPARDA CLI on your workspace and reads its JSON
output — no telemetry, no network calls of its own.