WebsitePublisher methods now throw on failure instead of silently
continuing. CLI entry points catch errors and exit(1) so GitHub
Actions correctly marks the job as failed.
- Refactor build.js: replace androidUpload/genericPublish/published with
unified WebsitePublisher class using Authorization Bearer auth
- New CLI commands: upload-changelog, release
- Update auto-update URL to /api/download/update (legacy compat on website)
- publish.yml: use CHANGELOG.md for GitHub Release body, replace
PUBLISH_KEY with UPLOAD_TOKEN + UPLOAD_URL
Remove self-description header, 38-line command listing, directory trees,
and standard Laravel patterns that Claude can infer from code.
Add 6 project-specific gotchas Claude would get wrong: non-REST routing
(InvokeController), custom response envelope (Base::retSuccess/retError),
AbstractModel::createInstance, Doo::userId auth, manual validation (no
FormRequest), and Swoole Task (not Laravel Queue).
122 lines → 46 lines.
New endpoint POST api/dialog/msg/send_ai_assistant sends messages
as the AI assistant identity (userid=-1). Supports both dialog_id
(direct) and task_id (with auto-creation) parameters.
Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Supports optional `with_extend` query parameter (comma-separated).
When `project_name` or `column_name` is included, the API returns
these fields inline with each task via eager loading.
Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
1. 描述建议：AI 返回 Markdown，前端用 MarkdownConver 转 HTML
2. 负责人推荐：排除已分配的任务成员
3. 解析负责人推荐时去重，防止 AI 返回重复用户
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
App\Models\Apps -> App\Module\Apps
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
The regex pattern (\w+) didn't match 'ai-apply' or 'ai-dismiss' because
\w doesn't include hyphens, causing all AI suggestion buttons to fail.
Fix by handling AI links before the regex match using startsWith().
Remove dead switch cases that were never reached.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
- Add STATUS_APPLIED and STATUS_DISMISSED constants to model
- Add markApplied() and markDismissed() methods
- Update event status after apply/dismiss actions (prevent duplicate ops)
- Validate related_task_id exists and user has permission
- Filter empty or overly long subtask names before creation
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Security fixes:
- Add escapeUserInput() to prevent Prompt injection via user input
- Validate msgId belongs to dialogId in updateMessageStatus()
- Add type parameter whitelist validation in ai-apply/ai-dismiss
- Add event record validation in task__ai_dismiss
Robustness fixes:
- Use atomic update for markProcessing to prevent concurrent processing
- Add subtask count limit check before creation (max 50)
- Disable similar task feature until vector search is implemented
- Fix Promise anti-pattern in frontend actions
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
