v1.7.0
sergeyzwezdin/hubav1.7.0Sep 7, 2026by umputun
Upgrading?
AI Summary
This release updates the default Codex model and improves logging/archiving, including fixes for Windows path handling and OAuth token persistence.
Key Highlights
- Updated default codex model to `gpt-5.6-sol`
- Archive completed progress logs instead of truncating them
- Fixed Windows path handling to use forward slashes in plan files
- Improved OAuth token persistence from container to host
- Improved detection of Claude limit patterns
New Features
- New default Codex model configuration
- Enhanced logging and plan archiving
Full Release Notes
**New Features** - Archive completed progress logs instead of truncating them #449 @umputun **Improved** - Update default codex model to gpt-5.6-sol and effort to high #421 @umputun Explicit `codex_model` / `codex_reasoning_effort` values in your own config are preserved on upgrade - edit them or comment them out to pick up the new defaults. - Stop pointing users at gpt-5.5 for codex 76910af - Describe Windows support as best effort 09fe05e - Refresh dependencies and fix the e2e playwright install #434 @paskal - Bump github.com/go-pkgz/notify from 1.4.0 to 1.5.0 #456 @dependabot - Bump github.com/stretchr/testify from 1.11.1 to 1.12.0 #432 @dependabot - Bump actions/setup-go from 6 to 7 #416 @dependabot **Fixed** - Match claude limit patterns by diagnostic provenance #455 @yeroo - Return repo-relative paths with forward slashes so the plan file excludes itself on Windows #454 @dimi4ik - Archive the completed plan on the feature branch #451 @umputun - Detect codex "Selected model is at capacity" as a limit pattern #448 @umputun - Let worktree mode start with a dirty source checkout #445 @umputun - Serve the worktree's plan copy to the dashboard #442 @umputun - Surface a failed plan archive instead of reporting a clean success #441 @umputun - Restrict the completed-plan archive commit to the plan paths #437 @umputun - Stop review agents from spawning subagents of their own #428 @umputun - Detect claude individual spend limit in default patterns #426 @umputun - Tighten signal rules in review prompts #424 @umputun - Enumerate API Error codes in default claude_error_patterns #423 @umputun - Pass max thinking level through to pi #433 @asalimonov - Persist OAuth token refreshes from container to host #412 @umputun - Harden OAuth credential bind mounts and keep older wrappers working 51f0d25 - Keep authenticated status in claude diagnostics and add weekly limit defaults 84100f4
Related Repositories
openclaw/openclaw★ 392,706
The AI that really does things. Any OS. Any Platform. The lobster way. 🦞
nilbuild/developer-roadmap★ 370,075
Interactive roadmaps, guides and other educational content to help developers grow in their careers.
morluto/rea★ 358,379
Reverse engineer anything with agents, from app behavior down to native binaries.
anomalyco/opencode★ 219,726
The open source coding agent.
n8n-io/n8n★ 208,261
Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.