[ca] Update OpenAI Codex CLI from 0.89.0 to 0.91.0

[github-actions]

Updated pkg/constants/constants.go with new version. Requires running make recompile to regenerate workflow lock files.

Update Summary

• Current Version: 0.89.0
• Latest Version: 0.91.0
• Intermediate Versions: 0.90.0
• Breaking Changes: No
• Update Priority: Medium
• Release Timeline:
  • 0.89.0: January 22, 2026
  • 0.90.0: January 25, 2026
  • 0.91.0: January 25, 2026

Key Features (0.90.0)

Major New Features:

• Network Sandbox Proxy with policy enforcement for better control over outbound network access
• Connectors Support via app server and MCP integration (initial phase)
• Collaboration Mode (Beta) shipped in TUI with improved plan-to-execute handoff
• Ephemeral Threads with improved collaboration tool provenance metadata
• WebSocket Proxy Support for proxy configuration

Enhanced Commands:

• /permissions command added with shorter approval set (maintains /approvals compatibility)
• /skill UI to enable or disable individual skills
• Improved slash-command selection prioritizing exact/prefix matches over fuzzy matches

App Server Improvements:

• Support for thread/read and filtering archived threads in thread/list
• Layered config.toml resolution for clients
• config/read can compute effective config from given cwd
• Release artifacts include stable URL for published config schema

Breaking Changes

None detected. All changes are additive features and fixes.

Bug Fixes (0.90.0)

• Fixed multiline quoted argument parsing in execution policies
• --yolo flag now skips git repository validation
• Enhanced resume functionality with out-of-order event handling and working directory prompts
• Corrected backspace behavior for text elements
• Improved configuration error messaging
• Default model selection respects filtered presets
• Prevented tilde expansion from escaping HOME on paths like ~//...
• TUI turn timing now resets between assistant messages

Chores (0.91.0)

• Reduced maximum allowed sub-agents from unlimited to 6 to tighten resource usage and improve guardrails in agent fan-out behavior
  • This is a constraint change that may affect workflows with heavy parallelization
  • Pull Request: chore: half max number of sub-agents openai/codex#9861

View Full Changelog

Version 0.91.0 (January 25, 2026)

Changes:

• Reduced the maximum allowed number of sub-agents to 6 to tighten resource usage and guardrails in agent fan-out behavior - chore: half max number of sub-agents openai/codex#9861

Contributors: @jif-oai

Full Changelog: openai/codex@rust-v0.90.0...rust-v0.91.0

---

Version 0.90.0 (January 25, 2026)

New Features:

• Network Sandbox Proxy with policy enforcement - feat: introducing a network sandbox proxy openai/codex#8442
• Connectors support via app server and MCP integration - [connectors] Support connectors part 1 - App server & MCP openai/codex#9667
• Collaboration Mode (Beta) in TUI - feat: tui beta for collab openai/codex#9690, TUI: prompt to implement plan and switch to Execute openai/codex#9712, Have a coding mode and only show coding and plan openai/codex#9802, Mark collab as beta openai/codex#9834
• Ephemeral threads with improved metadata - feat: ephemeral threads openai/codex#9765, feat: add thread spawn source for collab tools openai/codex#9769
• WebSocket proxy support - feat: support proxy for ws connection openai/codex#9719
• Stricter agent limitations on multi-agent operations

Bug Fixes:

• Fixed multiline quoted argument parsing - Fix execpolicy parsing for multiline quoted args openai/codex#9565
• --yolo flag skips git validation - fix(exec): skip git repo check when --yolo flag is used openai/codex#9590
• Enhanced resume functionality - Fix resume picker when user event appears after head openai/codex#9512, Ask for cwd choice when resuming session from different cwd openai/codex#9731
• Corrected backspace behavior - Prevent backspace from removing a text element when the cursor is at the element’s left edge openai/codex#9630
• Improved config error messaging - Another round of improvements for config error messages openai/codex#9746
• Fixed default model selection - Select default model from filtered presets openai/codex#9782

Documentation:

• Fixed typo in collaboration prompt template - Fix typo in experimental_prompt.md openai/codex#9716
• Updated bundled model catalog - Update models.json openai/codex#9726
• GitHub Actions updated for Node 24 - Upgrade GitHub Actions for Node 24 compatibility openai/codex#9722

---

Version 0.89.0 (January 22, 2026)

New Features:

• Added /permissions command - feat(tui) /permissions flow openai/codex#9561
• Added /skill UI - Add UI for skill enable/disable. openai/codex#9627
• Enhanced slash-command selection - feat: better sorting of shell commands openai/codex#9629
• App server thread/read support - feat(app-server): thread/read API openai/codex#9569, feat(app-server): support archived threads in thread/list openai/codex#9571
• Layered config.toml resolution - Add layered config.toml support to app server openai/codex#9510
• Config schema in release artifacts - feat: publish config schema on release openai/codex#9572

Bug Fixes:

• Fixed tilde expansion - fix: Fix tilde expansion to avoid absolute-path escape openai/codex#9621
• Fixed TUI turn timing - fix(tui) turn timing incremental openai/codex#9599

Documentation:

• Updated MCP subcommand docs - docs: fix outdated MCP subcommands documentation openai/codex#9622
• Refreshed skills/list README - Update skills/list protocol readme openai/codex#9623

Chores:

• Removed TUI2 experiment - feat(tui): retire the tui2 experiment openai/codex#9640

CLI Discovery

No CLI changes detected between versions 0.89.0 and 0.91.0. The --help output is identical across all three versions (0.89.0, 0.90.0, 0.91.0).

All changes are internal improvements to:

• Network sandbox proxy implementation
• MCP integration and connectors
• Collaboration mode features
• Bug fixes and stability improvements

Impact Assessment

• Risk: Low
• Affects:
  • Workflows using heavy parallelization (sub-agent limit reduced to 6 in v0.91.0)
  • Internal network access patterns (new sandbox proxy policies)
  • Users interested in the new Collaboration Mode beta features
• Testing Notes: The sub-agent limit reduction to 6 is the only constraint change that might affect existing workflows

Recommendations

Update Priority: Medium

Testing Strategy:

1. Verify workflows with parallel agent execution respect the new 6 sub-agent limit
2. Test network access patterns if using network sandbox features
3. Optional: Explore new Collaboration Mode beta features for improved planning workflows

Rollout Plan:

1. ✅ COMPLETED: Updated pkg/constants/constants.go (DefaultCodexVersion: "0.89.0" → "0.91.0")
2. REQUIRED: Run make recompile to regenerate workflow lock files
3. Test workflows with multiple parallel agents to verify 6 sub-agent limit
4. Monitor for any issues with network sandbox proxy if applicable

Package Links

• NPM Package: https://www.npmjs.com/package/`@openai/codex`
• Repository: https://github.com/openai/codex
• Release Notes: https://github.com/openai/codex/releases
• v0.91.0 Release: https://github.com/openai/codex/releases/tag/rust-v0.91.0
• v0.90.0 Release: https://github.com/openai/codex/releases/tag/rust-v0.90.0
• v0.89.0 Release: https://github.com/openai/codex/releases/tag/rust-v0.89.0

Action Required

Maintainer must run: make recompile after merging this change to regenerate workflow lock files (.github/workflows/*.lock.yml).

References:

• Workflow Run §21337980631

AI generated by CLI Version Checker

[github-actions]

🍪 Issue Monster has assigned this to Copilot!

I've identified this issue as a good candidate for automated resolution and assigned it to the Copilot agent.

The Copilot agent will analyze the issue and create a pull request with the fix.

Om nom nom! 🍪

AI generated by Issue Monster