Expose queued and active processing state for project sessions
- Dominant language
- Shell
- Stars
- 11.2k
- Forks
- 1.9k
- Avg merge
- 14h 16m
- Merged PRs (30d)
- 6
Description
### Describe the feature or problem you'd like to solve
Project-session coordination APIs do not expose whether a child session is actively processing or still has queued user messages. A parent can receive a completion handoff and see static branch/diff metadata from `get_session`, yet remain unable to prove the child is drained before merging its pull request.
This creates a cross-session merge race: if the parent previously sent a follow-up that is still queued, the child can wake after the parent merges, process that message, and push a trailing commit to the already-merged branch.
### Proposed solution
Expose explicit processing and queue state in `get_session` and `list_sessions_and_chats`, for example:
- `is_processing: boolean`
- `queued_messages_count: number`
- the current/last processed message identifier or a monotonic drain generation
Provide a native way for a coordinator to establish a merge barrier: wait until the child reports no active turn and no queued messages, then send a final stop/ready request and verify its acknowledgment belongs to the latest queue generation.
This would make project-session orchestration safe without relying only on conversational conventions or guessing from `updated_at`.
### Example prompts or workflows
1. A parent asks a child to implement a PR, sends two review fixes, then checks `is_processing == false` and `queued_messages_count == 0` before merging.
2. A parent sends “Make no further changes; reply READY_FOR_MERGE,” then verifies the reply drained the latest queue generation before running `gh pr merge`.
3. An orchestration dashboard shows which child sessions are idle, actively processing, or waiting on queued input.
4. A parent avoids archiving a child that still has pending messages even though its latest task handoff is complete.
### Additional context
Observed workflow: a child reported a validated PR head; the parent independently reviewed and merged it. A previously queued follow-up then woke the child, which produced and pushed a test-only commit after merge. The child detected the race and stopped, but the parent had no API field that could have prevented it.
Related queue/session-state reports do not expose a coordinator merge barrier:
- [Messages submitted while background subagents are running get stranded in the Queued UI region](https://github.com/github/copilot-cli/issues/3344)
- [Queued user messages and system notifications delivered out of send order](https://github.com/github/copilot-cli/issues/3517)
- [Project sessions hide context-exhausted and interrupted turn causes](https://github.com/github/copilot-cli/issues/4144)
- [Plan approval in autopilot+fleet mode triggers a dispatch ordering mismatch](https://github.com/github/copilot-cli/issues/2833)
Contributor guide
Research direction
Start by tracing the get_session and list_sessions_and_chats API paths, then read the related queue and session-state reports linked in the issue. Define the state and generation semantics needed for a coordinator to verify that a child has no active or queued work before merging, and document what acknowledgment makes the barrier complete.
Written by the indexing model from the issue text.
Assessment
- Domain
- api, backend-api-design, cli
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100