Keep the reasoning
Plans and findings stay in local JSON files, with optional HTML views. Sprint journals and /compound preserve decisions and solutions for later work.
Aguara detects threats in your software supply chain, from malicious packages to agent instructions. nanostack gives AI coding agents a complete workflow, from scoping a change to checking it and preparing a release.
Both projects are available under Apache 2.0.

An attack can start before your application runs.
A package update can introduce credential theft through an install script. A repository can bring CI workflows or agent settings that grant more access than you intended.
Aguara checks known malicious packages and analyzes code, configuration and instructions for these threats. Scans run locally and deterministically; signed threat-intelligence updates are opt-in.
Explore Aguara on GitHubAvailable as a CLI and GitHub Action, with JSON and SARIF output for your review workflow.
Check supported packages and lockfiles against known malicious-package records.
aguara check .Combine package checks with source inspection, including install scripts that download and execute another payload.
aguara audit . --ciInspect instructions, MCP configuration and agent settings without trusting the repository’s own suppressions.
aguara scan . --project-policy ignoreA malicious dependency can steal credentials during installation. Aguara checks known malicious-package records against supported lockfiles and installed packages.
npm, pnpm, Yarn and Bun lockfiles; installed packages.
Installed packages, .pth files and supported caches.
Gemfile.lock
Package lockfiles and C#, F# or Visual Basic project files.
go.sum and go.mod
Cargo.lock, for public-registry packages.
composer.lock
pom.xml and Gradle lockfiles.
Matching depends on the ecosystem: npm supports version ranges; others use exact versions and applicable all-version records. This is malicious-package detection, not general CVE scanning.
See detailed coverageGive your coding agent a process you can inspect.
Start with the problem, agree on the scope, then build and check the change. nanostack connects those steps through saved plans, findings and test results, so the next session can pick up the work.
Use the built-in sprint or add your team’s own checks. Skills are readable files, and the workflow runs locally with your coding agent.
Get started with nanostackAdapters for Claude Code, Cursor, OpenAI Codex, OpenCode and Gemini CLI.
Clarify the problem and the smallest useful change.
Save a plan with scope, affected files, risks and checks.
Implement the agreed change.
Review the code and catch changes outside the plan.
Examine threats, vulnerabilities and proposed fixes.
Exercise the browser, API or CLI and record results.
Check release evidence, prepare the PR and verify CI.
Production deployment remains an explicit decision.
Plans and findings stay in local JSON files, with optional HTML views. Sprint journals and /compound preserve decisions and solutions for later work.
/conductor manages task dependencies and handoffs between agent sessions. It groups compatible phases for parallel work and supports resuming an interrupted sprint.
The included custom-workflow example adds license review, privacy checks and release readiness before shipping. Your own skills can use the same evidence and phase system.
Build a custom workflowOn Claude Code, hooks can block risky commands and commits missing required review, security or QA evidence. Other adapters provide guidance rather than the same blocking guarantees. Check the host enforcement matrix.
Aguara powers Signal’s inspection engine. nanostack is a separate development workflow. Control and Cloud handle supported runtime policy and evidence workflows. Their capabilities and releases are distinct.