diff --git a/frontend/packages/agent/src/stats.html b/frontend/packages/agent/src/stats.html
index 52076b78..72fcf32a 100644
--- a/frontend/packages/agent/src/stats.html
+++ b/frontend/packages/agent/src/stats.html
@@ -33,6 +33,10 @@
     <div class="stats-card wide"><h3>token cost per bucket (sum across inferences)</h3><div class="chart-wrap"><canvas id="chart-cost"></canvas></div></div>
     <div class="stats-card wide"><h3>turns by model per bucket — model drives token cost</h3><div class="chart-wrap"><canvas id="chart-model"></canvas></div></div>
     <div class="stats-card"><h3>top tools</h3><div class="chart-wrap"><canvas id="chart-tools"></canvas></div></div>
+    <!-- "favorite tools": most-run shell commands. Hidden until the
+         bash_commands capture (hive-bash-mcp) has recorded data, so the
+         card never shows a permanently-empty doughnut. -->
+    <div class="stats-card" id="card-bash" hidden><h3>favorite tools (bash)</h3><div class="chart-wrap"><canvas id="chart-bash"></canvas></div></div>
     <div class="stats-card"><h3>wake source mix</h3><div class="chart-wrap"><canvas id="chart-wake"></canvas></div></div>
     <div class="stats-card"><h3>result mix</h3><div class="chart-wrap"><canvas id="chart-result"></canvas></div></div>
     <div class="stats-card wide"><h3>result trend per bucket — errors / rate-limits / compactions over time</h3><div class="chart-wrap"><canvas id="chart-result-trend"></canvas></div></div>
diff --git a/frontend/packages/agent/src/stats.js b/frontend/packages/agent/src/stats.js
index 08b2c23e..01c5ee92 100644
--- a/frontend/packages/agent/src/stats.js
+++ b/frontend/packages/agent/src/stats.js
@@ -340,8 +340,26 @@ window.Chart = Chart;
     });
   }
 
+  // "favorite tools" doughnut: most-run shell commands. The capture
+  // (hive-bash-mcp -> bash_commands table) lands separately, so until
+  // there's data we hide the whole card rather than show an empty
+  // doughnut. Runs independently of turn_count (a bash task is tied to
+  // a turn, but we don't want to couple the two reads).
+  function renderBashCard(s) {
+    const card = document.getElementById('card-bash');
+    const items = s.bash_breakdown || [];
+    if (!items.length) {
+      if (card) card.hidden = true;
+      destroy('chart-bash');
+      return;
+    }
+    if (card) card.hidden = false;
+    renderKeyCount('chart-bash', items, 'no bash commands');
+  }
+
   function render(s) {
     renderSummary(s);
+    renderBashCard(s);
     if (s.turn_count === 0) {
       paintEmpty('chart-turns', 'no turns in window');
       paintEmpty('chart-duration', 'no turns in window');
diff --git a/hive-ag3nt/src/stats.rs b/hive-ag3nt/src/stats.rs
index f9fdf14e..ff28d916 100644
--- a/hive-ag3nt/src/stats.rs
+++ b/hive-ag3nt/src/stats.rs
@@ -90,6 +90,13 @@ pub struct Snapshot {
     pub buckets: Vec<Bucket>,
     /// Top tools by call count across the window. Capped to 10.
     pub tool_breakdown: Vec<KeyCount>,
+    /// Top shell commands ("favorite tools") by invocation count across
+    /// the window, capped to 10. Normalised command heads recorded per
+    /// bash task into the `bash_commands` table by hive-bash-mcp. Empty
+    /// until that capture lands (or on any agent that hasn't run a bash
+    /// task) — the table is created lazily by the writer, so a read
+    /// before the first insert returns an empty list, not an error.
+    pub bash_breakdown: Vec<KeyCount>,
     pub wake_mix: Vec<KeyCount>,
     pub result_mix: Vec<KeyCount>,
     /// Distinct models seen in the window, sorted. Each bucket's
@@ -176,6 +183,7 @@ fn empty_snapshot(window: Window) -> Snapshot {
         turn_count: 0,
         buckets,
         tool_breakdown: Vec::new(),
+        bash_breakdown: Vec::new(),
         wake_mix: Vec::new(),
         result_mix: Vec::new(),
         models: Vec::new(),
@@ -276,6 +284,7 @@ fn snapshot(path: &Path, window: Window) -> Result<Snapshot> {
         turn_count,
         buckets,
         tool_breakdown: top_n(tool_totals, 10),
+        bash_breakdown: read_bash_breakdown(&conn, from).unwrap_or_default(),
         wake_mix: top_n(wake_totals, 20),
         result_mix: top_n(result_totals, 20),
         models,
@@ -284,6 +293,36 @@ fn snapshot(path: &Path, window: Window) -> Result<Snapshot> {
     })
 }
 
+/// Aggregate the top shell-command heads ("favorite tools") over
+/// `[from, now]` from the `bash_commands` table — one row per bash task
+/// (`ts INTEGER NOT NULL, head TEXT NOT NULL`), written by hive-bash-mcp.
+///
+/// Returns `Err` (which the caller maps to an empty list) when the
+/// table doesn't exist yet — the writer creates it lazily on first
+/// insert, so any agent that hasn't run a bash task since the capture
+/// shipped simply has no table. Decoupling it this way means the read
+/// side is inert-until-data and needs no schema coordination here.
+fn read_bash_breakdown(conn: &Connection, from: i64) -> rusqlite::Result<Vec<KeyCount>> {
+    let mut stmt = conn.prepare(
+        "SELECT head, COUNT(*) AS n
+         FROM bash_commands
+         WHERE ts >= ?1
+         GROUP BY head",
+    )?;
+    let rows = stmt.query_map([from], |row| {
+        Ok((
+            row.get::<_, String>(0)?,
+            u64_from_i64(row.get::<_, i64>(1)?),
+        ))
+    })?;
+    let mut totals: HashMap<String, u64> = HashMap::new();
+    for r in rows {
+        let (head, n) = r?;
+        *totals.entry(head).or_insert(0) += n;
+    }
+    Ok(top_n(totals, 10))
+}
+
 struct Row {
     started_at: i64,
     duration_ms: i64,
@@ -581,4 +620,60 @@ mod tests {
         assert_eq!(s.bucket_seconds, 86_400);
         assert!(s.buckets.len() >= 7);
     }
+
+    /// `bash_breakdown` degrades gracefully when the `bash_commands`
+    /// table hasn't been created yet (the capture side hasn't shipped /
+    /// run on this agent). A `seed_db` DB has no such table, so the read
+    /// must yield an empty list rather than erroring the whole snapshot.
+    #[test]
+    fn bash_breakdown_empty_without_table() {
+        let db = tmp_db();
+        let _ = std::fs::remove_file(&db);
+        seed_db(&db, &[(now_secs() - 100, 1000, "opus", "recv", "ok", "{}")]);
+        let s = snapshot(&db, Window::Day).unwrap();
+        assert!(s.bash_breakdown.is_empty());
+    }
+
+    /// With a populated `bash_commands` table, `bash_breakdown` rolls up
+    /// per-head counts (busiest first) and respects the window cutoff.
+    #[test]
+    fn bash_breakdown_aggregates_heads() {
+        let db = tmp_db();
+        let _ = std::fs::remove_file(&db);
+        seed_db(&db, &[]);
+        let now = now_secs();
+        let conn = Connection::open(&db).unwrap();
+        conn.execute_batch("CREATE TABLE bash_commands (ts INTEGER NOT NULL, head TEXT NOT NULL);")
+            .unwrap();
+        // 3x cargo + 2x git inside the window, 1x rg outside it.
+        for (ts, head) in [
+            (now - 100, "cargo"),
+            (now - 200, "cargo"),
+            (now - 300, "cargo"),
+            (now - 400, "git"),
+            (now - 500, "git"),
+            (now - (2 * 24 * 3600), "rg"), // older than the 24h window
+        ] {
+            conn.execute(
+                "INSERT INTO bash_commands (ts, head) VALUES (?1, ?2)",
+                params![ts, head],
+            )
+            .unwrap();
+        }
+        let s = snapshot(&db, Window::Day).unwrap();
+        let map: HashMap<_, _> = s
+            .bash_breakdown
+            .iter()
+            .map(|kc| (kc.key.clone(), kc.count))
+            .collect();
+        assert_eq!(map.get("cargo").copied(), Some(3));
+        assert_eq!(map.get("git").copied(), Some(2));
+        // `rg` fell outside the 24h window — excluded.
+        assert_eq!(map.get("rg").copied(), None);
+        // top_n orders busiest first.
+        assert_eq!(
+            s.bash_breakdown.first().map(|kc| kc.key.as_str()),
+            Some("cargo")
+        );
+    }
 }