Skip to content

v0.1.0

Choose a tag to compare

@smshagor-dev smshagor-dev released this 14 Nov 04:34
· 3 commits to main since this release
0aca629

CodeTwin ML v0.1.0 — Build for testing

Release channel: Early Testing Preview (Pre-release)
Date: 14 November 2026

Testing release, not a production-ready declaration. These notes describe the first testable feature set visible in the repository. They do not certify that an installer has been published, that all CI checks pass, or that full physical end-to-end testing is complete.

Overview

CodeTwin ML is a local-first desktop workspace for understanding, reviewing, and improving software projects. It combines project indexing, a persistent software digital twin, security analysis, authorized web testing, on-device AI-assisted review, and evidence-aware remediation.

This first testing release focuses on validating real project imports, scanning coverage, agent execution, stability, and the Windows desktop experience.

What's included

Project workspace and code intelligence

  • Project import, source indexing, persisted file/symbol/dependency information, and navigation through the codebase.
  • Progress and coverage evidence for long-running imports and scans.
  • Work on resumable project indexing and transaction-safe persistence.
  • Project-scoped AI chat, agent-based file review, a line-by-line terminal view, and annotated findings.

Security and web testing

  • Code-quality, dependency, secret, database, runtime, and security review capabilities across supported inputs.
  • Authorized web-target discovery with a persistent SQLite-backed queue and resumable read-only GET processing.
  • Scan progress, receipts, and explicit reporting of incomplete, skipped, or inconclusive coverage.
  • Redaction and evidence-handling improvements for shareable reports.

Local AI and datasets

  • On-device agent setup and model-readiness checks before agent-dependent scanning.
  • Optional local cybersecurity reference datasets with download/status controls.
  • Dataset-assisted context for agent chat and file review.
  • No bundled model is assumed; some features require a separately configured local model.

Fixes and packaging work

  • Repairs to streamed project import transactions, database integrity handling, and build integration.
  • Windows app and NSIS installer branding using the CodeTwin ML logo.
  • Release collection script that gathers locally built installers under release/v0.1.0/windows/x64/ and writes metadata/SHA256SUMS.txt.

What testers should validate

  1. Installation and startup: launch the Windows build, verify branding, navigation, setup, and clean uninstall.
  2. Project import: test small and large repositories, interrupt/restart imports, and verify no silent omission or false completion.
  3. AI agent: configure an on-device model, test chat and line-by-line reviews, and confirm clear behavior if the model is unavailable.
  4. Security analysis: inspect findings, paths, line numbers, severity, supporting evidence, and false-positive handling.
  5. Authorized web scans: test only systems you own or have explicit permission to assess; validate progress, pause/resume, receipts, and coverage gaps.
  6. Recovery: exercise cancellations, process restarts, corrupt inputs, network interruptions, and retry behavior.
  7. Remediation: verify approvals, backups, hash checks, rollback, and independent retesting before calling a fix verified.
  8. Performance: record project size, file/line counts, scan duration, memory use, and reproducible test conditions.

Known limitations and release readiness

  • Not certified for production: end-to-end Windows desktop/installer validation and complete current-head CI evidence were not confirmed for these notes.
  • Installer availability: the repository has NSIS packaging configuration and a collector script, but a downloadable GitHub Release asset was not verified.
  • Signing: the production release pipeline requires signing credentials and clean-machine install/uninstall checks; do not assume the testing build is signed.
  • Agent dependency: AI reviews require a working locally configured model and may vary with model capability.
  • Coverage is not a vulnerability guarantee: partial scans, inconclusive results, and unsupported inputs must not be interpreted as clean or secure.
  • Web testing must remain authorized and scoped. This preview does not establish that any target is vulnerability-free.

Feedback and bug reports

When reporting a problem, include the source commit, OS version, project size, steps to reproduce, expected versus actual behavior, and sanitized logs or screenshots. Never share secrets, proprietary source, or unredacted vulnerability details in public issues. Use the repository's private security reporting guidance for sensitive findings.

Project: CodeTwin-ML
Release status: Testing preview — feedback and validation requested.