Code UX 0.9.6
Code UX 0.9.6
Code UX 0.9.6 improves agent continuity and execution visibility: persistent skills are now discoverable in agent context and management UI, every supported local CLI provider has normalized live invocation parsing, imported work retains its full planning description, and sprint recovery handles restart and CI-repair edges more reliably. See the complete promotion in release PR #2850.
Release Snapshot
| Item | Details |
|---|---|
| Version | 0.9.6 |
| Release commit | b685b1a4b4a1243b7042f927dda84f5194fc3abd |
| Release PR | #2850 |
| Included dev sync | #2851 |
| Scope | 218 files changed across agent skills/settings, provider telemetry, issue import, sprint reliability, model catalog, and documentation |
What’s New
Persistent Skills and Agent Settings
Persistent storage now supplies an authenticated, scoped skill inventory and concise descriptions to invoked agents.
- Added persistent-skill disclosures on agent storage interactions and a dedicated persistent-storage management experience.
- Reorganized Agents settings into focused production-ready panels and reconciled deleted storage attachments safely.
- Preserved non-destructive interactions while making attached skills easier to discover and manage.
Provider Invocation Visibility
Every supported local CLI provider now has normalized live invocation parsing for richer execution visibility.
- Added or hardened parsing for Gemini, Codex, Claude Code, Qwen, OpenCode, and Antigravity.
- Persists normalized user, assistant, reasoning, tool-call, and tool-result turns for invocation inspection.
- Improved provider usage accounting and transcript resilience across coding and non-coding invocation types.
Imported Work Context
Imported issue descriptions now remain available throughout planning and execution.
- Retains full source descriptions through search, persistence, MCP imports, planning context, and dashboard updates.
- Prevents empty refresh payloads from destructively replacing an already captured description.
Sprint Recovery and Catalog Updates
Sprint execution is more resilient to restart and CI-repair edge cases.
- Recovers completed provider work after restart and resumes eligible task work without duplicate execution.
- Improves exhausted CI-repair visibility and prevents restart races from concealing actionable attention.
- Refreshes the bundled models.dev catalog and aligns canonical and published documentation.
Fixes
Agent and Provider Reliability
- Fixed persistent-skill context being unavailable to an invoked agent despite configured storage.
- Fixed stale storage attachment state and incomplete settings interactions after storage removal.
- Fixed incomplete transcript and usage parsing for provider-specific and non-coding invocation paths.
Planning and Orchestration
- Fixed imported issue bodies being omitted downstream and vulnerable to destructive empty updates.
- Fixed restart-recovery and CI-repair timing races that could leave recoverable work paused or hidden.
- Fixed main-merge gating and git-status recovery edge cases through additional orchestration coverage.
Validation
| Check | Result |
|---|---|
| Typecheck & Lint | Passed |
| Backend Tests & Coverage | Passed — 397 files; 5,068 tests passed, 7 skipped |
| Dashboard Tests | Passed |
| Security Audit | Passed |
| Playwright E2E | Passed — all 18 Linux/macOS/Windows shards |
| CodeQL Analysis | Passed |
Additional local validation:
pnpm run ci
pnpm run check:docs-web
CODE_UX_SKIP_RELEASE_INSTALL_BUILD=1 node scripts/verify-release-install.mjs
git diff --checkGitHub validation also passed release policy, package-install smoke, five-page Docs smoke, Linux Docker orchestration, macOS/Windows Electron orchestration, and unsigned Linux/macOS/Windows desktop release-candidate packages.
Known Follow-Up
- Desktop artifacts remain intentionally unsigned and may show operating-system trust prompts until signing is configured.