OpenSquilla — Token-Efficient AI Agent

Evidence shows: README details installation telemetry, data collection scope (explicitly lists what is sent and not sent), privacy switch (OPENSQUILLA_PRIVACY_DISABLE_NETWORK_OBSERVABILITY), and uninstall options that preserve data. This reflects good practices in data flow transparency and sensitive data handling. However, there is no specific description of least privilege, user confirmation mechanisms (e.g., confirmation before tool execution), and external effects (like network requests) are described but lack user control details. For rollback, uninstall commands support --dry-run and data preservation, but no configuration or state rollback is mentioned. For source attribution, authors are listed as 'OpenSquilla contributors', but publisher is unverified and no clear contributor attribution or source traceability is provided. Therefore, data flow transparency scores 2, sensitive data handling scores 2, rollback scores 2, and others score 1 due to insufficient evidence or partial coverage.

Evidence shows: README and pyproject.toml have consistent version numbers (0.5.2), and CI configuration includes multiple test matrices, indicating good internal consistency. However, for dependency availability, the dependency list is broad and includes many optional extras, but no dependency locking or version pinning strategy is provided, which may affect reproducibility. For failure messages, README mentions some error handling (e.g., DLL load failed, libomp missing), but does not systematically describe error message design. Therefore, self-consistency scores 2, dependency availability and failure messages each score 1.

Evidence shows: README clearly lists target audiences (desktop users, terminal users, developers) and multiple usage scenarios (desktop install, quick terminal install, install from source, develop from source), and provides detailed installation path choices. For capability boundaries, documentation explains which features require additional extras (e.g., matrix, document-extras), but does not clearly state the Agent's capability limits. For trigger precision, no detailed description of command or tool trigger conditions. For environment fit, supports Windows, macOS, Linux, and provides platform-specific installation instructions. Therefore, audience and scenarios scores 2, capability boundaries scores 2, environment fit scores 2, trigger precision scores 1.

Evidence shows: README has a clear structure with sections for installation, configuration, running, uninstall, etc., indicating good information architecture. Installation notes are very detailed, covering multiple platforms and methods, scoring 3. For naming stability, project name and command (opensquilla) are consistent, but no stability guarantee for API or configuration is provided. For examples and FAQ, README provides multiple command examples but no FAQ. For known limitations, README mentions some platform-specific issues (e.g., Windows DLL, macOS libomp), but does not systematically list all limitations. License is Apache-2.0, file is complete, scoring 3. For versioning and changelog, README mentions version 0.5.2 but no detailed changelog. For maintenance responsibility, authors are listed as 'OpenSquilla contributors', but no clear maintainers or contribution guidelines. Therefore, information architecture scores 2, install notes scores 3, naming stability scores 1, examples and FAQ scores 2, known limitations scores 1, license scores 3, versioning and changelog scores 1, maintenance responsibility scores 1.

Evidence shows: README describes the product's core value (token efficiency, intelligent routing, persistent memory, secure sandbox) and provides multiple usage methods (CLI, Web UI, chat channels), indicating good output usability. For marginal value, the product claims higher intelligence density at the same budget, but lacks independent verification. For cost-benefit, documentation mentions token efficiency but provides no specific cost data. Therefore, output usability scores 2, marginal value scores 2, cost-benefit scores 2.

Evidence shows: Some claims in README (e.g., 'Token-efficient', 'higher intelligence density') lack specific data or references, so claim traceability scores 1. For cross-source corroboration, no external validation or independent test results are provided, scoring 1. For fact-inference separation, documentation describes product features as facts but does not clearly distinguish which are inferences or promises, scoring 1.