Local trust layer for AI-generated code changes.
AvaTrust watches selected workspace files while you code, flags risky trust patterns early, and helps you review changes before they ship.
AI-assisted coding can produce changes that look reasonable but quietly introduce trust risks:
- preview or debug bypasses
- browser-reachable secret exposure
- client-side token leaks
- unsafe HTML rendering
- sensitive routes without explicit auth checks
AvaTrust focuses on these high-signal patterns without uploading your code anywhere.
- You open a trusted workspace in a supported editor.
- AvaTrust watches active files locally.
- When it sees a risky change, it adds:
- inline diagnostics
- Problems panel findings
- Quick Fix actions
- You review the finding, apply the safer pattern yourself, and re-check the file.
Main commands:
AvaTrust: Review Current FileAvaTrust: Review Current Changes
AvaTrust v1 is built for VS Code-compatible editors.
Runtime smoke checks passed on this machine in:
- Visual Studio Code
- Visual Studio Code Insiders
- Cursor
- Windsurf
- VSCodium
JetBrains support is intentionally out of scope for this product track right now.
Codex Desktop is supported today as a companion review surface through the AvaTrust Codex plugin.
Current Codex support is companion-style rather than inline diagnostics:
- review the current file after edits
- review changed files in a workspace
- get a chat-friendly trust summary
- keep a local
last-review.mdandlast-review.jsonartifact
It should still be described as a Codex companion, not as a native inline editor overlay.
This first version ships as a local-only Free Preview.
Included in Free Preview:
- local live trust hints
- current file review
- current changes review
- safer fix guidance
Preview limits:
20 live trust finding sessions / month40 review actions / month
Limit behavior:
- a live trust finding session is one meaningful batch of new findings in the current file or change set, not one count per individual issue
- a review action is an explicit user action such as
Review Current File,Review Current Changes, or opening a safer-fix guidance flow
Planned paid tiers:
Solo - coming soonPro - coming soon
AvaTrust is designed to be local-first.
- no code upload
- no file contents sent to us
- no diffs sent to us
- no scan payload leaves your environment
- selected workspace scope only
Optional anonymous telemetry is planned, but the product should always work without telemetry.
Telemetry, when enabled, will be limited to product-behavior events such as:
- extension activation
- active day
- review opened
- quick fix opened
- extension version
AvaTrust will not collect:
- source code
- file contents
- diffs
- secrets
- scan bodies
- report bodies
This early version is intentionally focused on high-signal trust issues. It is not trying to replace a full semantic reviewer, a full security platform, or a general-purpose AI coding assistant.
It is best at catching:
- trust-sensitive bypasses
- exposed secret-like values
- client-side token leaks
- obviously risky frontend trust patterns
- missing auth signals on sensitive routes
MunaTrust and AvaTrust are related but intentionally separate.
MunaTrust:
- broader scanner / review / ship-gate product
- existing marketplace distribution
AvaTrust:
- dedicated live trust-layer product
- local-first workspace watcher behavior
- focused on early detection and safer-fix guidance
The repository also includes AvaTrust Codex, a Codex companion plugin that reuses AvaTrust review logic in Codex workflows.
It is designed for:
- reviewing the current file from Codex
- reviewing changed files before a broader pass
- producing chat-ready findings
- leaving a local review artifact for visibility
Install from GitHub with Codex:
codex plugin marketplace add balkanbrs/avatrust --ref main --sparse .agents --sparse plugins/avatrust-codex
codex plugin add avatrust-codex@avatrustIf AvaTrust catches something important in your workflow, that is exactly the signal this product is built for.