story-kit: create 94_bug_stale_agent_state_persists_after_server_restart
This commit is contained in:
@@ -0,0 +1,27 @@
|
|||||||
|
---
|
||||||
|
name: "Stale agent state persists after server restart"
|
||||||
|
---
|
||||||
|
|
||||||
|
# Bug 94: Stale agent state persists after server restart
|
||||||
|
|
||||||
|
## Description
|
||||||
|
|
||||||
|
The agent pool is in-memory. When the server restarts, agent processes are gone but their records linger as pending entries that never get cleaned up. The frontend shows stale agent assignments (e.g. coder-1 still assigned to a story that finished long ago). Pipeline advancement also fails because the server was not running when the agent exited, so stories get stuck in 2_current instead of advancing to 3_qa.
|
||||||
|
|
||||||
|
## How to Reproduce
|
||||||
|
|
||||||
|
1. Start the server and launch a coder on a story\n2. Let the coder complete its work and commit\n3. Restart the server\n4. Observe the web UI — the story still shows the coder assigned, pipeline stage did not advance
|
||||||
|
|
||||||
|
## Actual Result
|
||||||
|
|
||||||
|
Stale agent assignments persist in the UI. Stories get stuck in their current pipeline stage because the server missed the agent process exit event.
|
||||||
|
|
||||||
|
## Expected Result
|
||||||
|
|
||||||
|
On startup, the server should either (a) clear all agent records since the processes are gone, or (b) reconcile state by checking which worktrees have committed work and running gates/advancing the pipeline accordingly.
|
||||||
|
|
||||||
|
## Acceptance Criteria
|
||||||
|
|
||||||
|
- [ ] On server startup, stale agent records from a previous session are cleared
|
||||||
|
- [ ] Stories with completed work in their worktree are detected and advanced through the pipeline on startup
|
||||||
|
- [ ] No phantom agent lozenges appear in the UI after a server restart
|
||||||
Reference in New Issue
Block a user