Grux for Mac

The README describes on-demand, optional permissions and a filesystem path allowlist, denylist, limits, redaction, and audit log. It also acknowledges that the app is unsandboxed and can run real shell commands, so least privilege, confirmation, and rollback each score 2. Data flows are detailed for the Keychain, model providers, meeting audio, and local logs, earning 3. Direct dependencies are listed, but the supplied material does not establish their current vulnerability status, so dependency security scores 1. Security reporting channels and a maintainer email are given, but publisher identity is unverified, so source attribution scores 1.

The README, CI configuration, and launcher tests give concrete descriptions of tool counts, test floors, failure paths, and known CI gaps; some product claims cannot be independently checked from the supplied files, so self-consistency scores 2. Dependencies and the minimum toolchain are described, but availability and compatibility are not evidenced here, so dependency availability scores 2. The CLI offers status, doctor, why, and exit-code information, and tests cover several error messages; clear recovery for every failure is not established, so failure messages score 2.

The target environment is specifically Apple silicon on macOS 14 or later, with local Ollama and hosted-model scenarios described. The feature set is broad, but audience-specific guidance is limited, so audience and scenarios score 2. The permission table, feature tiers, and security boundaries describe capabilities and acknowledge the lack of sandboxing and rough Labs features, earning 2. Permissions are requested when needed and refusal disables the corresponding feature, but confirmation details for other high-impact actions are incomplete, so trigger precision scores 2. Platform, toolchain, and model choices are explained, but only Mac is supported, so environment fit scores 2.

The README has a contents list and sections for installation, permissions, feature tiers, and building; it is long and the supplied copy is truncated, so information architecture scores 2. DMG, Homebrew, CLI, checksum, and source-build steps are explicit, earning 3. Product version and feature names are described, but naming rules are not, so naming stability scores 2. There are many FAQ links and command examples, but the linked answers cannot be checked from the supplied material, so examples and FAQ score 2. The security policy documents limitations around redirects, DNS, and the redactor, and the README labels Labs as rough, earning 2. The MIT license text is complete, earning 3. The README identifies version 3.0 and CI and security files describe release support, but a full change history is absent from the supplied material, so versioning and changelog score 2. The security email, response timelines, and supported-version policy identify maintenance responsibilities; the solo-project context and unverified identity limit that evidence, so maintenance responsibility scores 2.

The CLI supports JSON reads and status, doctor, and why commands; launcher tests cover argument forwarding and terminal output. Actual task outcomes are not verified by this static review, so output usability scores 2. The unified interface, system integrations, and reusable Swift modules suggest useful value, balanced by the broad permissions and complexity, so marginal value scores 2. MIT licensing and local-model use reduce cost; hosted models require user-paid accounts, while hardware, permissions, and setup impose costs, so cost-benefit scores 2.

Many installation, security, and feature claims point to specific files, configuration, or test descriptions, but some important claims remain README assertions, so claim traceability scores 2. The README, CI, security policy, and launcher tests corroborate some claims; execution results and all implementation files are not supplied, so cross-source corroboration scores 2. The docs distinguish core features from Labs and acknowledge several limitations, but do not consistently label fact, commitment, and inference, so fact-inference separation scores 2.