wedged-run fix: pre-approve claude perms + verbose step log + reaper

Root cause: claude CLI in the per-topic research container runs as
uid=0(root). ZeroClaw's claude_cli provider passes
--dangerously-skip-permissions which Claude CLI rejects under root
for security — so the CLI hangs waiting for interactive permission
approval that never arrives, hitting the 600s provider timeout with
zero step records journaled.

Three-part fix:

1. Claude settings bind-mount (research_container.rs):
   Optional CLAWMATES_CLAUDE_SETTINGS_PATH env — when set, mount the
   host file at /root/.claude/settings.json (read-only) in every
   spawned team container. deploy/claude-settings.json ships the
   canonical config (permissions.defaultMode = bypassPermissions +
   hasCompletedOnboarding). CLI accepts requests immediately with no
   --dangerously-skip-permissions flag needed.

2. Verbose per-step log line (topology_worker.rs):
   Every checkpoint now writes to stderr:
     topology_worker::step run_id=X step=N node=Y role=Z phase=W
       output_bytes=B tokens=T gated=G
   Visible in docker logs clawmates_server_1 — gives us live
   'topology is flowing' signal without opening the canvas, and
   makes it obvious when a topology_kind is skipping stages it
   shouldn't.

3. Stuck-container reaper (topology_worker.rs):
   New 60s-tick loop reap_stuck_runs: for any research topology_run
   older than 15 min with zero checkpoint.records, docker-stop its
   container and mark the run failed with a diagnostic error. Only
   reaps research-bound runs (non-research runs don't own a
   container). The existing 180s stale-checkpoint requeuer stays
   in place for other failure modes.

Deploy: gw-04 needs
  ln -sf /path/to/repo/deploy/claude-settings.json /opt/clawmates/claude-settings.json
  CLAWMATES_CLAUDE_SETTINGS_PATH=/opt/clawmates/claude-settings.json
  in the server env, plus the timeout lowered from 600 -> 120 in
  compose. Both handled in the deploy step outside this commit.

crates/cm-api/src/research_container.rs

@@ -222,7 +222,7 @@ pub async fn spawn(
     // disabling pairing here is safe.
     prewrite_daemon_config(state_host_path)?;
 
-    let mounts = vec![
+    let mut mounts = vec![
         Mount {
             target: Some("/workspace/repo".into()),
             source: Some(repo_host_path.to_string_lossy().into_owned()),
@@ -239,6 +239,29 @@ pub async fn spawn(
         },
     ];
 
+    // 2026-07-15 (post-diagnosis): claude CLI in-container runs as
+    // uid=0 and refuses --dangerously-skip-permissions under root for
+    // security. Without pre-approved permissions the CLI hangs waiting
+    // for interactive approval → the whole run stalls until the 600s
+    // provider timeout. Bind-mount a settings.json with
+    // permissions.defaultMode=bypassPermissions so the CLI accepts
+    // requests immediately.
+    //
+    // Path via env so deployments can swap in a different settings
+    // file (e.g. a workspace-specific one) without a code change. When
+    // unset the mount is skipped (dev backwards compat).
+    if let Ok(claude_settings_path) = std::env::var("CLAWMATES_CLAUDE_SETTINGS_PATH") {
+        if !claude_settings_path.is_empty() {
+            mounts.push(Mount {
+                target: Some("/root/.claude/settings.json".into()),
+                source: Some(claude_settings_path),
+                typ: Some(MountTypeEnum::BIND),
+                read_only: Some(true),
+                ..Default::default()
+            });
+        }
+    }
+
     let host_config = HostConfig {
         mounts: Some(mounts),
         network_mode: Some(team_network()),

crates/cm-api/src/topology_worker.rs

@@ -25,9 +25,28 @@ use crate::topology_exec::ZeroClawDriveExecutor;
 /// Requeue a `running` job whose worker hasn't checkpointed within this window.
 const STALE_AFTER_SECS: f64 = 180.0;
 
+/// Maximum age a `running` run may spend WITHOUT journaling any step
+/// records before the reaper kills its container and fails it. 15 min
+/// is generous: healthy first-step latency is typically 5–60s; anything
+/// past this is a stuck container (usually a wedged provider CLI).
+const REAP_STUCK_AFTER_SECS: i64 = 15 * 60;
+
 /// Spawn the durable topology job worker. Polls for queued jobs every `poll`
 /// interval; runs each to completion (or failure), checkpointing per step.
 pub fn spawn(pool: PgPool, runtime: cm_runtime::Runtime, poll: Duration) {
+    // Fire the stuck-container reaper on its own cadence — checking
+    // once a minute is plenty and keeps this off the hot claim loop.
+    let reaper_pool = pool.clone();
+    tokio::spawn(async move {
+        let mut ticker = tokio::time::interval(Duration::from_secs(60));
+        loop {
+            ticker.tick().await;
+            if let Err(e) = reap_stuck_runs(&reaper_pool).await {
+                eprintln!("topology_worker::reaper: reap failed: {e}");
+            }
+        }
+    });
+
     tokio::spawn(async move {
         loop {
             // Recover jobs orphaned by a dead worker before claiming new ones.
@@ -47,6 +66,65 @@ pub fn spawn(pool: PgPool, runtime: cm_runtime::Runtime, poll: Duration) {
     });
 }
 
+/// Kill research team containers whose `running` topology_run has been
+/// alive past [`REAP_STUCK_AFTER_SECS`] without journaling a single
+/// step record. Marks the run `failed` with a diagnostic error so the
+/// user sees WHY instead of an infinitely-spinning pipeline.
+///
+/// Only reaps runs bound to a research topic — non-research runs (raw
+/// API-driven topology runs) don't own a container so there's nothing
+/// to kill; they're left to the existing stale-checkpoint requeuer.
+async fn reap_stuck_runs(pool: &PgPool) -> Result<(), sqlx::Error> {
+    use sqlx::Row;
+    let rows: Vec<sqlx::postgres::PgRow> = sqlx::query(
+        "SELECT id, research_topic_id
+           FROM topology_runs
+          WHERE status = 'running'
+            AND research_topic_id IS NOT NULL
+            AND created_at < now() - make_interval(secs => $1::float)
+            AND coalesce(jsonb_array_length(coalesce(checkpoint->'records', '[]'::jsonb)), 0) = 0",
+    )
+    .bind(REAP_STUCK_AFTER_SECS as f64)
+    .fetch_all(pool)
+    .await?;
+
+    if rows.is_empty() {
+        return Ok(());
+    }
+
+    // Best-effort docker cleanup; even if the container is already gone
+    // (crashed, manually killed), we still want to mark the run failed.
+    let docker = crate::research_container::connect().ok();
+
+    for row in rows {
+        let id: Uuid = row.get("id");
+        let topic_id: Uuid = row.get("research_topic_id");
+        let container = crate::research_container::container_name_for(topic_id);
+        eprintln!(
+            "topology_worker::reaper: reaping stuck run run_id={} topic_id={} container={} (no step records after {}s)",
+            id, topic_id, container, REAP_STUCK_AFTER_SECS,
+        );
+        if let Some(d) = &docker {
+            let _ = d
+                .stop_container(
+                    &container,
+                    None::<bollard::query_parameters::StopContainerOptions>,
+                )
+                .await;
+        }
+        let _ = cm_db::repo::topology_runs::fail(
+            pool,
+            id,
+            &format!(
+                "reaped: no step records after {}s (container {} stopped)",
+                REAP_STUCK_AFTER_SECS, container
+            ),
+        )
+        .await;
+    }
+    Ok(())
+}
+
 /// Drive one claimed job to a terminal state, persisting checkpoints as it goes.
 async fn run_job(
     pool: &PgPool,
@@ -473,6 +551,31 @@ async fn drive<E: TurnExecutor>(
     execute_resumable(graph, task, executor, progress, move |snap| {
         let pool = pool_cb.clone();
         async move {
+            // 2026-07-15: verbose per-step trace so `docker logs
+            // clawmates_server_1` shows which topology node just fired,
+            // its role, output size, tokens, and gated-action count.
+            // Cheap (one info! per step) and gives us the missing
+            // "topology is flowing" signal without needing to open the
+            // canvas.
+            if let Some(last) = snap.records.last() {
+                let phase = match &last.phase {
+                    cm_orchestrator::StepPhase::Plan => "plan",
+                    cm_orchestrator::StepPhase::Work => "work",
+                    cm_orchestrator::StepPhase::Synth => "synth",
+                    cm_orchestrator::StepPhase::Aggregate => "aggregate",
+                };
+                eprintln!(
+                    "topology_worker::step run_id={} step={} node={} role={} phase={} output_bytes={} tokens={} gated={}",
+                    id,
+                    snap.completed,
+                    last.node_id,
+                    last.role,
+                    phase,
+                    last.output.len(),
+                    last.tokens,
+                    last.gated.len(),
+                );
+            }
             // Best-effort checkpoint: a failed write just means we re-run the
             // step on resume (idempotent — topology turns are pure reads here).
             if let Ok(v) = serde_json::to_value(&snap) {

deploy/claude-settings.json

@@ -0,0 +1,6 @@
+{
+  "hasCompletedOnboarding": true,
+  "permissions": {
+    "defaultMode": "bypassPermissions"
+  }
+}