fix.

Automated AST-guided repository audit and multi-agent code repair.

We turn raw LLM-generated code into solid, test-verified engineering. Syntax-aware indexing finds the hotspots, isolated agent teams fix them, and real E2E suites prove the result before anything is committed.

How it works — technical details

AST-aware indexing: tree-sitter parses JS/TS/Swift and splits code strictly on syntax boundaries (classes, functions, hooks) up to 100 lines — no naive character splitting, no token-heavy greps.
Graph dependencies: millisecond dependency queries give agents exact call hierarchies and edge risk flags (unhandled async promises, missing ESM extensions, UI/logic monoliths).
Multi-agent tmux ADE: isolated Developer, Tester, and Orchestrator sessions coordinate over stdin/tmux pipe-pane; the Tester runs E2E suites blind, reporting only pass/fail to prevent prompt contamination.
Payment-gated execution: a 50% pre-payment confirmation creates an isolated git worktree and launches the autonomous repair pipeline; delivery is a clean, test-verified PR.

fix.

Start an audit

Audit summary