huskies: merge 1052

server/src/crdt_state/mod.rs

@@ -55,8 +55,8 @@ pub use types::{
 pub use write::{
     bump_retry_count, migrate_legacy_stage_strings, migrate_merge_job, migrate_names_from_slugs,
     migrate_node_claims_to_agent_claims, migrate_story_ids_to_numeric, name_from_story_id,
-    set_agent, set_depends_on, set_epic, set_item_type, set_name, set_plan_state, set_qa_mode,
-    set_resume_to, set_resume_to_raw, set_retry_count, write_item,
+    purge_done_stage_merge_jobs, set_agent, set_depends_on, set_epic, set_item_type, set_name,
+    set_plan_state, set_qa_mode, set_resume_to, set_resume_to_raw, set_retry_count, write_item,
 };
 
 #[cfg(test)]

server/src/crdt_state/write/migrations.rs

@@ -650,6 +650,61 @@ pub fn migrate_merge_job(db_path: &std::path::Path) {
     slog!("[crdt] Migrated {count} MergeJob entries to typed MergeResult format");
 }
 
+/// Delete MergeJob CRDT entries for stories that have reached a terminal stage
+/// (Done, Archived, Abandoned, Superseded, Rejected).
+///
+/// Pre-1036 code left stale MergeJob entries after a story recovered from a
+/// merge failure and transitioned to Done.  Those entries caused the gateway
+/// UI to mislabel recovered stories as "FAILED" because `load_pipeline_state`
+/// unconditionally read the MergeJob error field for every story.  This
+/// migration removes the orphaned entries so the state is clean on the next
+/// server start.
+///
+/// Running this migration repeatedly is safe — tombstoned entries are filtered
+/// out by the read path, so subsequent calls are no-ops.
+pub fn purge_done_stage_merge_jobs() {
+    use crate::pipeline_state::Stage;
+
+    let Some(jobs) = crate::crdt_state::read_all_merge_jobs() else {
+        return;
+    };
+    if jobs.is_empty() {
+        return;
+    }
+
+    // Collect IDs of every story currently in a terminal stage.
+    let terminal_ids: std::collections::HashSet<String> = crate::pipeline_state::read_all_typed()
+        .into_iter()
+        .filter(|item| {
+            matches!(
+                item.stage,
+                Stage::Done { .. }
+                    | Stage::Archived { .. }
+                    | Stage::Abandoned { .. }
+                    | Stage::Superseded { .. }
+                    | Stage::Rejected { .. }
+            )
+        })
+        .map(|item| item.story_id.0.clone())
+        .collect();
+
+    let to_delete: Vec<String> = jobs
+        .into_iter()
+        .filter(|j| terminal_ids.contains(&j.story_id))
+        .map(|j| j.story_id)
+        .collect();
+
+    if to_delete.is_empty() {
+        return;
+    }
+
+    let count = to_delete.len();
+    for story_id in &to_delete {
+        crate::crdt_state::delete_merge_job(story_id);
+    }
+    slog!("[crdt] Purged {count} stale MergeJob entries for terminal-stage stories");
+}
+
 #[cfg(test)]
 mod merge_job_migration_tests {
     use super::super::super::state::init_for_test;

server/src/crdt_state/write/mod.rs

@@ -19,4 +19,5 @@ pub use item::write_item_str;
 pub use migrations::{
     migrate_legacy_stage_strings, migrate_merge_job, migrate_names_from_slugs,
     migrate_node_claims_to_agent_claims, migrate_story_ids_to_numeric, name_from_story_id,
+    purge_done_stage_merge_jobs,
 };

server/src/http/workflow/pipeline.rs

@@ -116,7 +116,9 @@ pub fn load_pipeline_state(ctx: &AppContext) -> Result<PipelineState, String> {
 
         // Story 945: review_hold is `Stage::ReviewHold`; qa_mode and epic_id
         // come from typed CRDT registers. merge_failure detail lives on the
-        // MergeJob CRDT entry (same as status_tools.rs).
+        // MergeJob CRDT entry, but only surfaces for items currently in a
+        // failure stage — Done/terminal items must never inherit stale errors
+        // from a previously-failed merge attempt (story 1052).
         let view = crate::crdt_state::read_item(sid);
         let review_hold = if matches!(item.stage, Stage::ReviewHold { .. }) {
             Some(true)
@@ -125,7 +127,14 @@ pub fn load_pipeline_state(ctx: &AppContext) -> Result<PipelineState, String> {
         };
         let qa = view.as_ref().and_then(|v| v.qa_mode());
         let epic_id = view.as_ref().and_then(|v| v.epic());
-        let merge_failure = crate::crdt_state::read_merge_job(sid).and_then(|j| j.error);
+        let merge_failure = if matches!(
+            item.stage,
+            Stage::MergeFailure { .. } | Stage::MergeFailureFinal { .. }
+        ) {
+            crate::crdt_state::read_merge_job(sid).and_then(|j| j.error)
+        } else {
+            None
+        };
 
         let story = UpcomingStory {
             story_id: sid.clone(),

server/src/service/pipeline/mod.rs

@@ -47,7 +47,7 @@ pub fn aggregate_pipeline_counts(pipeline: &Value) -> Value {
             .map(|f| !f.is_null() && f != "")
             .unwrap_or(false);
 
-        if is_blocked || has_merge_failure {
+        if (is_blocked || has_merge_failure) && stage != "done" {
             let story_id = item
                 .get("story_id")
                 .and_then(|s| s.as_str())

server/src/startup/project.rs

@@ -167,6 +167,9 @@ pub(crate) async fn init_subsystems(app_state: &Arc<SessionState>, cwd: &Path) {
             crdt_state::migrate_merge_job(db_path);
             // Story 1009: drop legacy node-hex claims that can't be converted to AgentName.
             crdt_state::migrate_node_claims_to_agent_claims();
+            // Story 1052: remove stale MergeJob entries for terminal-stage
+            // stories so they can never cause "FAILED" labels in the UI.
+            crdt_state::purge_done_stage_merge_jobs();
         }
     }
 }