v1.22.0

gentle-ai v1.22.0

This release brings a really nice quality jump across Claude, OpenSpec, prompt behavior, and install reliability.

Highlights

• Claude now gets SDD slash commands

  • managed command installation under ~/.claude/commands
  • full SDD command set for Claude workflows
  • agent-aware command asset selection

• Gentleman responses are now shorter by default

  • less over-explaining
  • less ritualized scaffolding
  • better just-answer-the-thing behavior
  • still keeps the Gentleman tone

• Better OpenSpec documentation

  • clarified how openspec/config.yaml works in practice with gentle-ai
  • aligned docs with the real OpenSpec-based workflow we use

• Improved language behavior

  • fixed cases where the Gentleman persona could drift into Spanish during English conversations

• Backup safety improvements

  • excludes ~/.gemini/antigravity/tmp/ from backups

• More clarity around custom

  • clarified that custom means keeping the user's existing setup and persona unmanaged
  • this was a semantics and UX fix, not a runtime behavior change

Reliability improvements

• improved OpenCode SDD orchestrator prompt guidance
• hardened Engram release download behavior when GitHub restricts token-based API access
• fixed Qwen integration to use direct settings injection instead of unsupported engram setup qwen-code
• improved E2E reliability across Arch and Fedora paths

Community contributors in this release

• Jose Manuel Delgado Trueba (@DelgadoTrueba) for the OpenSpec config documentation contribution in PR #344

Thanks

A bunch of these fixes came directly from community reports and follow-up debugging on real workflows. Thanks for the sharp eyes and patience while we chased the weird edge cases.