hive-bash-mcp: explain the status when returning it, not up front

Mirrors #4333's split on the sibling hive-subagent-mcp surface (mara's
ruling there, quoted on #4336: "dont explain all the possible states
that can be returned in the tool description. instead explain the
status when returning it").

status_explanation() carries what each TaskStatus means and what to do
about it, appended to every format_task answer. The status tool's
description shrinks to what it's for; a caller only ever sees one
state at a time and now learns what it means without cross-referencing
anything else.

hive-bash-mcp/src/mcp.rs

@@ -30,9 +30,35 @@ fn status_str(status: &TaskStatus) -> &'static str {
     }
 }
 
+/// What a state means and what the caller should do about it — carried on
+/// every `format_task` answer instead of the tool description enumerating
+/// all six up front, so a caller who only ever sees one state at a time
+/// still learns what it means without reading anything else.
+fn status_explanation(status: &TaskStatus) -> &'static str {
+    match status {
+        TaskStatus::Pending => "queued, not started yet",
+        TaskStatus::Running => {
+            "still going — check back later, or pass wait_seconds to block briefly"
+        }
+        TaskStatus::Done => "finished normally",
+        TaskStatus::TimedOut => "exceeded timeout_secs and was killed",
+        TaskStatus::Interrupted => {
+            "the daemon restarted mid-task — the process is gone, no result to report"
+        }
+        TaskStatus::Killed => {
+            "killed on request via `kill` — a signal was sent to the whole process group"
+        }
+    }
+}
+
 /// Format a `TaskFile` as the human-readable status string claude sees.
 fn format_task(task: &TaskFile) -> String {
-    let mut out = format!("task `{}`: status={}", task.id, status_str(&task.status));
+    let mut out = format!(
+        "task `{}`: status={} ({})",
+        task.id,
+        status_str(&task.status),
+        status_explanation(&task.status)
+    );
 
     if let Some(code) = task.exit_code {
         let _ = write!(out, ", exit={code}");
@@ -189,9 +215,9 @@ impl BashMcp {
     }
 
     #[tool(
-        description = "Check the status of a background bash task by its ID (from `run`). \
-            Returns the current status (pending/running/done/timed_out/interrupted/killed), exit \
-            code if finished, and a tail of stdout/stderr. Full output lives in \
+        description = "Check the status of a background bash task by its ID (from `run`). The \
+            answer names the state it found and what to do about it, plus the exit code if \
+            finished and a tail of stdout/stderr. Full output lives in \
             `harness/bash-tasks/<id>.out` / `.err`. \
             Pass `wait_seconds` (capped at 30) to wait inline for the task to finish: when the \
             task finishes within the window the full status is returned immediately. Useful to \
@@ -332,4 +358,31 @@ mod status_hint_tests {
             assert!(out.contains("status=done"));
         });
     }
+
+    #[test]
+    fn every_status_answer_names_its_state_and_the_next_move() {
+        // Pins the split this module's doc describes: the tool description
+        // says what the tool is for, and each individual answer — not a
+        // list the caller has to cross-reference — says what state it got
+        // and what to do about it. One state's explanation text is
+        // distinct enough from the others that a copy-paste regression
+        // (e.g. `Killed` picking up `TimedOut`'s wording) would fail this.
+        with_harness_dir(|| {
+            for status in [
+                TaskStatus::Pending,
+                TaskStatus::Running,
+                TaskStatus::Done,
+                TaskStatus::TimedOut,
+                TaskStatus::Interrupted,
+                TaskStatus::Killed,
+            ] {
+                let explanation = super::status_explanation(&status);
+                let out = format_task(&task(status.clone()));
+                assert!(
+                    out.contains(explanation),
+                    "status={status:?} explanation missing from: {out}"
+                );
+            }
+        });
+    }
 }