Integuru v0

The evidence shows that the tool reads HAR and cookie files, supplies request information and cookies to a cloud LLM, and generates code that can call platform internal endpoints. Input files constrain the scope, but the materials show no collection minimization, pre-execution confirmation, secret redaction, or rollback mechanism, so most criteria lose points. The README mentions local storage and LLM use but does not explain whether request data is sent to the model or how credentials are handled. Dependencies have version ranges but no lockfile or security review evidence. The code and README attribute the product to Integuru v0; publisher identity and ongoing maintenance responsibility are unclear.

The README describes graph construction, code generation, and CLI options, and the main entry point broadly matches that workflow. However, the README gives a default max_steps of 20 while main.py uses 15, and the provided source shows no specific failure feedback for the optional --generate-code path. The dependency list and CI configuration provide limited availability evidence; CI only declares pytest, and the sampled tests use mocks, so static evidence cannot establish the real platform workflow.

The README targets users who want integration code for platforms without official APIs and explains browser capture, prompts, and input variable scenarios. The CLI supports model, file path, and step customization. Boundaries include input variables currently being supported only for graph generation and the requirement that users log in and complete 2FA first. There is no systematic account of platform compatibility, authorization prerequisites, or unsupported actions, so this does not receive full marks.

The README covers features, setup, usage, tests, CI, 2FA, and privacy, with a CLI example. Setup specifies a Python/Poetry environment, OpenAI API key, and model. The full AGPL-3.0 LICENSE supports full marks for licensing. The project is labeled v0, the package version is 0.1.0, and a newer version is mentioned on an external site, but no changelog is shown. A contact email and contribution invitation are present, while responsibility and the update path remain limited.

The product has a specific goal: infer a dependency graph from browser network requests and generate runnable Python request code. The README gives a utility bill example and workflow, supporting practical value for integrations without official APIs. The output can be used as code, but there is no evidence of generated-code security review, execution preview, action confirmation, or recovery. Model calls, browser capture, and credential preparation add setup cost, so the scores are moderate to low.

The README workflow and feature claims partly correspond to the CLI options, main entry point, and agent logic for URL and dynamic parameter identification. The privacy statement only mentions local files and a cloud LLM, without adequately explaining the data flow; the CI declaration cannot statically prove tests passed. The evidence distinguishes some documented facts from code behavior, but marketing claims, privacy details, and actual effectiveness lack corroboration across sources, so points are deducted.