Cloud project session can start without a repository checkout
- 主要语言
- Shell
- 星标
- 11.2k
- 派生
- 1.9k
- 平均合并
- 14 小时 16 分钟
- 30 天内合并 PR
- 6
描述
## Problem
A coordinated cloud project session can report successful creation even though it has no repository checkout, workspace path, or runnable terminal. The session then cannot perform the implementation or validation work it was created for, and the creator receives no provisioning error.
## Reproduction
1. Create a coordinated cloud session for a configured repository project and an explicit base branch.
2. Allow the session to start automatically with a kickoff prompt that requires repository commands.
3. Inspect the session metadata and terminal.
Observed in session `8079b444-7187-470c-854f-ffa1f5ad74f2`:
- Session type was `cloud`, but its path was empty.
- The process working directory was `/`.
- No Git checkout was available.
- The native terminal surface twice returned `Terminal not found or not running`.
- The session exhausted its attempt budget without reaching the requested work.
## Expected behavior
- A cloud session created for a repository project starts with that repository and requested ref checked out in a runnable development environment.
- If repository/ref resolution or environment provisioning fails, session creation fails visibly with an actionable error.
- A repository-targeted cloud session never silently falls back to repository-less task state.
## Acceptance criteria
- Cloud-task creation validates that a repository-targeted request has a complete owner and repository name before creating the task.
- A partial repository object, including branch without owner/name, is rejected rather than converted to the empty-repository sentinel.
- Provisioning failure is surfaced through project-session status and the kickoff turn.
- Tests cover repository-targeted cloud creation, explicit repository-less creation, partial repository metadata, and terminal/environment startup failure.
- Programmatic project-session creation receives the same Codespaces-scope and provisioning errors as interactive CLI flows.
Investigation notes
- Runtime triage found that cloud-task metadata creation can omit the repository when either owner or name is missing, then serialize an empty owner/name sentinel while preserving the branch.
- The empty-repository sentinel is intended for explicitly repository-less tasks, not degraded repository-targeted sessions.
- A separate Codespaces OAuth-scope gap may affect the managed-environment path, but server telemetry is needed to determine which path this session used.
- Useful telemetry: the task-creation repository payload, whether an environment ID was assigned, and the provisioning failure/status transition for the session ID above.
Related issues:
- https://github.com/github/copilot-cli/issues/4054
- https://github.com/github/copilot-cli/issues/2655
- https://github.com/github/copilot-cli/issues/4144
贡献指南
调研方向
从云任务创建和项目会话配置的入口点开始,然后跟踪仓库元数据在环境和终端启动过程中的处理。运行或补充针对以下情况的覆盖:以仓库为目标、明确不使用仓库、元数据不完整以及配置失败;通过可见的创建错误或可运行的已检出会话来表明完成。
由索引模型根据 Issue 内容生成。
评估
- 技术栈
- github, shell
- 领域
- backend, cli, cloud
- Issue 类型
- 缺陷
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 活跃度
- 冷清
- 描述清晰度
- 基本清楚
- 新手友好度
- 48/100