AutomatiQ Web Session Recorder

The README says recordings store request and response bodies, cookies, and credentials without redaction. It documents configuration, proxies, telemetry, and output paths, but shows no field-level access controls or redaction, so least privilege and sensitive-data handling score 1. Recording is started by a user tool call and opens a visible browser; no separate confirmation for consequential actions is shown, so confirmation scores 1. Telemetry fields and opt-out are explicit, and recording/model data flows are described, earning 2 for transparency. Dependencies have pinned or bounded versions, but the supplied material does not show vulnerability review or lockfile contents, so dependency security scores 1. Browser requests, MCP session-file writes, optional proxies, and model calls are described, but telemetry is enabled by default and recording sends requests to external sites, so external-effects control scores 1. Re-annotation backs up originals and config updates retain a .bak, earning 2 for rollback. pyproject and the license identify the author and copyright, earning 2 for attribution; the unverified registry identity neither adds nor removes points.

The README, configuration, and tool descriptions are broadly consistent about recording, waiting, status, and output. The alpha notice limits stability claims, so self-consistency scores 2. A Python minimum and dependency list are provided, but the lockfile contents and fallback for unavailable dependencies are not, so availability scores 1. The README describes initialization delays, crash reports, and re-annotation, while the supplied tests cover errors such as unknown sessions, missing clips, and an unconfigured vision model; failure messages score 2.

The project targets developers and MCP clients, with a clear scenario: turn user browsing into material for an Agent to build HTTP automation. Audience and scenarios score 2. Tools, vision models, proxies, telemetry, and output have defined roles; the Brave-only recorder and purpose of video analysis are stated, so capability boundaries score 2. MCP tools have names, parameters, and a recording-to-artifact workflow, earning 2 for trigger precision. Python versions, multiple client configurations, environment variables, config files, and a local model endpoint are documented. The Brave-only recorder and macOS screen-recording permission requirement limit portability, so environment fit scores 2.

The README is organized around workflow, installation, MCP setup, tools, configuration, and development; it is clear but has little FAQ material, so information architecture scores 2. It gives pip installation, Python requirements, first-run download behavior, and several client configurations, earning 2 for install notes. Package, MCP server, command, and project names are consistent, earning 2 for naming stability. It includes config examples and a tool workflow but no systematic FAQ, so examples and FAQ score 2. Alpha status, unredacted data, Brave requirements, first-run delay, and model configuration are disclosed, earning 2 for known limitations. Both pyproject and LICENSE specify MIT, with the full license text, earning 3. Version is dynamically read from a config module, but no changelog or release process is supplied, so versioning and changelog score 1. An author and contact are listed, but maintenance responsibility, response expectations, and update path are not clear, so that criterion scores 1.

The recording artifact includes a session README, network data, actions, and optional vision annotations, and clients are directed to read the README first; output usability scores 2. Converting visible browsing into reusable material for HTTP automation has a clear potential benefit, but claims of greater accuracy and speed lack comparative evidence in the supplied material, so marginal value scores 2. Browser download, optional vision model/API key, video capture, and telemetry impose setup, time, or possible cost; no cost estimate or measured benefit is provided, so cost-benefit scores 1.

Most product, configuration, telemetry, and limitation claims are traceable to the README and pyproject. Test files show error paths, isolation measures, and re-annotation behavior, but this review does not claim they were executed; claim traceability scores 2. The README, project metadata, workflows, and tests provide some corroboration, but no run results, release history, or independent source supports key benefit claims, so cross-source corroboration scores 1. Alpha status, telemetry fields, unredacted data, and the Brave restriction are stated as facts and distinguishable from performance and accuracy aspirations, earning 2 for fact-inference separation.