v0.3.1: Codex hook timeouts

• Codex caps its SessionEnd and Interrupt hooks at 3s and warned on every start. Masterwork now writes 3s for those two.
• Already connected Codex? The Sessions screen shows Repair. Click it, then trust the hooks again in Codex's /hooks (terminal Codex), since trust is tied to the hook definition.

v0.3.0: Claude Code or Codex, everywhere

Claude Code or Codex, everywhere

Every feature now works for Codex users as well as Claude Code users.

• Pick the assistant in the sidebar. Chat, project summaries, links, trigger guides, generality audits, simulations, diagrams, skill match and factory launches all run on Claude Code (claude -p) or Codex (codex exec, read-only sandbox). A chat carries its conversation over when you switch agents.
• Instructions has tabs for ~/.claude/CLAUDE.md and ~/.codex/AGENTS.md, and warns when AGENTS.override.md shadows it.
• Codex assets: custom agents in ~/.codex/agents/*.toml, Codex plugin skills, and skills switched off in config.toml. Shared ~/.agents/skills now count Codex as a loader, since Codex reads that folder itself.
• Catalog installs go to Claude Code, Codex or the shared folder. Updating a skill that was made generic no longer breaks it.
• Sessions and Analytics filter by agent, and skill links in a run open the right asset page. Codex hooks need trusting in /hooks after Connect.
• Factory: factory/run.py --agent codex. Codex reports no price, so max_tokens is the budget cap.

Upgrading an existing install applies migrations 0031 and 0032 on start.

v0.2.1

What's Changed

• Feat/skill catalog by @flieks in #1
• Feat/skill catalog by @flieks in #2
• fix(ci): backend tests no longer need a claude on the runner's PATH by @flieks in #3
• Pre-launch: site, disable/enable skills, upstream drift, Codex sessions, 0.2.1 by @flieks in #4

New Contributors

• @flieks made their first contribution in #1

Full Changelog: https://github.com/flieks/masterwork/commits/v0.2.1