v1.7.0

parruda/swarmv1.7.0Sep 7, 2026by umputun

AI Summary

This release focuses on improving stability and Windows compatibility while updating default AI models. It introduces a new feature to archive progress logs and fixes numerous issues related to plan archiving, token management, and limit detection.

Key Highlights

Archive completed progress logs instead of truncating them
Updated default Codex model to gpt-5.6-sol with high effort
Improved Windows support with repo-relative paths and best effort status
Enhanced OAuth token persistence and credential handling
Better detection of Claude and Codex API limits

New Features

Archive completed progress logs instead of truncating them

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