All releases

Post-task reviews, guided agent input, and serverless chat.

Agent_00

@jefking @MoltenBot000

  • Added configurable post-task review levels with one, three, or six review passes, progress stages for findings and fixes, and safeguards against reprocessing automated review comments.

  • Added optional merged-branch cleanup after pull request completion, including fork-aware deletion, base-branch protection, and keeping tasks visible when deletion fails.

  • Accepted explicit prompt instructions to leave a repository unchanged when no changes exist, avoiding false no-change failures.

Molten Bot

@shayan919293

  • Added in-chat forms for missing details and secure credentials, including saved session secrets, alternative authentication choices, and Codex device login.

  • Added a Molten-managed AWS Bedrock provider option and refreshed Claude model bundles across BigBrain quality tiers.

  • Routed portal HTTP, BigBrain, terminal, and event traffic through dedicated serverless API and WebSocket paths across local, QA, and production environments.

  • Updated app subscriptions to use product-scoped billing plans, checkout, and portal flows, with normalized plan status and pricing.

AI Augmented Engineering

@shayan919293

  • Aligned task filters with live workflow statuses, added failed and cancelled filtering, clarified restart and auto-refresh states, and moved advanced task options behind progressive disclosure.

  • Improved keyboard and screen-reader behavior across dialogs, task controls, onboarding checks, password fields, overview tiles, and theme contrast.

  • Added Coding LLM settings for switching between local models and OpenRouter, with secure API-key storage, model selection, and connection health checks.

  • Added planner questions for text, choices, confirmations, and secrets so jobs can pause for human input, store credentials safely, and resume with focused replanning.

  • Added corrective build retries, repeated-failure detection, clearer blocked-plan diagnostics, validation and secret preflights, and manual Jira ticket job submission.

  • Added model-matrix and repository-task benchmarks for comparing agent providers, models, costs, timing, and validation outcomes.