1.1.4
- Added one shared built-in task-writing standard for Codex and Claude Code, distributed to every user through Agent Integrations. It requires an overview, concrete steps, observable acceptance criteria, a realistic estimate and genuine references when available.
- All authenticated test-connection responses, including exported protected-config connections, return the same task\_writing\_standard for clients to discover.
- Replaced the title-only setup example with a complete project-scoped request shape. Both agents are told to replace placeholders, preserve existing task IDs/keys during repairs and avoid invented facts or outcomes.
- Existing integrations can reconnect/repair to refresh their generated instructions while preserving valid grants and unrelated project instructions. Existing minimal clients remain transport-compatible.
- Added checks that both generated agent files contain the same quality standard, complete valid examples and the same readable task sections, without duplicate rules after repair.