From 6092f7efbb97611226af71e0b1b66b1881733c4f Mon Sep 17 00:00:00 2001
From: dave <futurechimp@users.noreply.github.com>
Date: Tue, 28 Apr 2026 23:06:40 +0000
Subject: [PATCH] huskies: merge 822

---
 .../src/agents/pool/pipeline/advance/mod.rs   | 22 ++++-
 .../pool/pipeline/advance/tests_regression.rs | 86 +++++++++++++++++++
 2 files changed, 107 insertions(+), 1 deletion(-)

diff --git a/server/src/agents/pool/pipeline/advance/mod.rs b/server/src/agents/pool/pipeline/advance/mod.rs
index a7931c76..db1b868f 100644
--- a/server/src/agents/pool/pipeline/advance/mod.rs
+++ b/server/src/agents/pool/pipeline/advance/mod.rs
@@ -13,6 +13,26 @@ use tokio::sync::broadcast;
 use super::super::super::{CompletionReport, PipelineStage, agent_config_stage, pipeline_stage};
 use super::super::{AgentPool, StoryAgent};
 
+/// Maximum number of bytes of gate output to include in the failure context
+/// injected into the resumed session.  Keeps the injected message focused —
+/// the tail of the output (where errors appear) is always preserved.
+const MAX_GATE_OUTPUT_BYTES: usize = 8_000;
+
+/// Truncate gate output to [`MAX_GATE_OUTPUT_BYTES`], keeping the **tail**
+/// (where compiler errors and test failures are reported).
+fn truncate_gate_output(output: &str) -> &str {
+    if output.len() <= MAX_GATE_OUTPUT_BYTES {
+        return output;
+    }
+    let start = output.len() - MAX_GATE_OUTPUT_BYTES;
+    // Advance to the next valid UTF-8 char boundary.
+    let mut adjusted = start;
+    while !output.is_char_boundary(adjusted) {
+        adjusted += 1;
+    }
+    &output[adjusted..]
+}
+
 impl AgentPool {
     /// Pipeline advancement: after an agent completes, move the story to
     /// the next pipeline stage and start the appropriate agent.
@@ -197,7 +217,7 @@ impl AgentPool {
                             "\n\n---\n## Previous Attempt Failed\n\
                              The acceptance gates failed with the following output:\n{}\n\n\
                              Please review the failures above, fix the issues, and try again.",
-                            completion.gate_output
+                            truncate_gate_output(&completion.gate_output)
                         );
                         if let Err(e) = self
                             .start_agent(
diff --git a/server/src/agents/pool/pipeline/advance/tests_regression.rs b/server/src/agents/pool/pipeline/advance/tests_regression.rs
index 0eefd928..a7b79121 100644
--- a/server/src/agents/pool/pipeline/advance/tests_regression.rs
+++ b/server/src/agents/pool/pipeline/advance/tests_regression.rs
@@ -770,3 +770,89 @@ async fn gates_failed_with_test_evidence_and_committed_work_advances() {
         "run_tests evidence must be cleared after pipeline advance consumes it"
     );
 }
+
+// ── story 822: warm-resume coder on gate failure ──────────────────────────
+
+/// Story 822 / AC 1 & 4: when a coder fails gates and a prior session ID is
+/// provided, the pipeline re-spawns the coder so it can warm-resume the prior
+/// conversation with the failure context injected — rather than starting from
+/// scratch and re-reading the spec.
+///
+/// The test verifies:
+/// - The coder is re-spawned (Pending/Running) rather than blocked.
+/// - The retry counter is incremented (AC 3).
+#[tokio::test]
+async fn warm_resume_coder_on_gate_failure_with_session_id() {
+    use std::fs;
+
+    let tmp = tempfile::tempdir().unwrap();
+    let root = tmp.path();
+
+    // Set up a project with a fast no-op coder agent.
+    fs::create_dir_all(root.join(".huskies")).unwrap();
+    fs::write(
+        root.join(".huskies/project.toml"),
+        r#"
+max_retries = 3
+
+[[agent]]
+name = "coder-1"
+role = "Coder"
+command = "echo"
+args = ["noop"]
+prompt = "test prompt"
+stage = "coder"
+"#,
+    )
+    .unwrap();
+
+    crate::db::ensure_content_store();
+    crate::db::write_item_with_content(
+        "9950_story_warm_resume",
+        "2_current",
+        "---\nname: Warm Resume Test\n---\n",
+    );
+
+    let pool = AgentPool::new_test(3001);
+
+    // Simulate a coder failing gates.  A prior session ID is provided to
+    // trigger the warm-resume path (--resume <session_id>).
+    pool.run_pipeline_advance(
+        "9950_story_warm_resume",
+        "coder-1",
+        CompletionReport {
+            summary: "Tests failed".to_string(),
+            gates_passed: false,
+            gate_output: "error[E0308]: mismatched types\n --> src/lib.rs:5:10".to_string(),
+        },
+        Some(root.to_path_buf()),
+        None,
+        false,
+        Some("prior-session-abc123".to_string()),
+    )
+    .await;
+
+    // The coder must be re-spawned — Pending or Running.
+    let agents = pool.agents.lock().unwrap();
+    let coder_restarted = agents.values().any(|a| {
+        a.agent_name == "coder-1" && matches!(a.status, AgentStatus::Pending | AgentStatus::Running)
+    });
+    assert!(
+        coder_restarted,
+        "Coder must be re-spawned (warm-resumed) when gates fail and prior session ID provided. \
+         Pool: {:?}",
+        agents
+            .iter()
+            .map(|(k, a)| format!("{k}: {} ({})", a.agent_name, a.status))
+            .collect::<Vec<_>>()
+    );
+    drop(agents);
+
+    // Retry counter must have been incremented (AC 3).
+    let content = crate::db::read_content("9950_story_warm_resume")
+        .expect("story must exist in content store");
+    assert!(
+        content.contains("retry_count"),
+        "retry_count must be incremented after warm-resume: {content}"
+    );
+}