From 66f340a7a378429a1fb64d6cd055052ef49ca961 Mon Sep 17 00:00:00 2001
From: dave <futurechimp@users.noreply.github.com>
Date: Thu, 30 Apr 2026 18:19:01 +0000
Subject: [PATCH] fix: prune session_store on stdio abort, respawn cold
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The bug 882 abort-respawn safeguard caps consecutive crashes at 5 then
blocks the story — but the underlying stdio abort itself stays unfixed:
each respawn calls start_agent which reads session_store.json, finds the
prior session id, passes --resume to claude-code, and re-triggers the
same crash. Five identical respawns later, the story is blocked.

Now: when an abort+no-session exit triggers respawn, we first call
session_store::remove_sessions_for_story to drop every entry for the
story. The next spawn starts cold (no --resume), which avoids the
bloated stdio replay claude-code is choking on.

The function was already implemented but #[cfg(test)] only — promoted
to a non-test pub fn. Existing remove_sessions_for_story_cleans_up test
unchanged and still green.

Net effect: instead of "5 retries, then blocked", we get "1 abort, prune,
respawn cold, agent runs normally". The story can resume work without
losing its worktree state.
---
 server/src/agents/pool/start/spawn.rs | 11 ++++++++++-
 server/src/agents/session_store.rs    |  8 ++++++--
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/server/src/agents/pool/start/spawn.rs b/server/src/agents/pool/start/spawn.rs
index 43238e64..946d6b24 100644
--- a/server/src/agents/pool/start/spawn.rs
+++ b/server/src/agents/pool/start/spawn.rs
@@ -463,10 +463,19 @@ pub(super) async fn run_agent_spawn(
                         reason,
                     });
                 } else {
+                    // Prune session_store entries for this story so the next
+                    // spawn starts cold (no `--resume` flag). The crash likely
+                    // came from claude-code choking on the bloated stdio
+                    // replay; resuming again would re-trigger the same abort.
+                    crate::agents::session_store::remove_sessions_for_story(
+                        &project_root_clone,
+                        &sid,
+                    );
                     slog!(
                         "[agents] CLI crashed before session for '{sid}:{aname}' \
                          (abort respawn {count}/{ABORT_RESPAWN_CAP}). \
-                         Respawning without consuming a retry slot."
+                         Pruned session_store and respawning cold without \
+                         consuming a retry slot."
                     );
                     let agents_for_respawn = Arc::clone(&agents_ref);
                     let watcher_for_respawn = watcher_tx_clone.clone();
diff --git a/server/src/agents/session_store.rs b/server/src/agents/session_store.rs
index 0a3f2c03..717e556f 100644
--- a/server/src/agents/session_store.rs
+++ b/server/src/agents/session_store.rs
@@ -73,8 +73,12 @@ pub fn lookup_session(
     read_store(project_root).get(&key).cloned()
 }
 
-/// Remove all session entries for a story (called when a story reaches done/archived).
-#[cfg(test)]
+/// Remove all session entries for a story.
+///
+/// Called when the story reaches done/archived, OR when claude-code keeps
+/// crashing on session resume — in the latter case the next spawn must start
+/// cold (no `--resume` flag) so the bloated stdio replay doesn't re-trigger
+/// the same abort. See bug 882 follow-up.
 pub fn remove_sessions_for_story(project_root: &Path, story_id: &str) {
     let mut data = read_store(project_root);
     let prefix = format!("{story_id}:");