@runuai/host 0.8.19 → 0.8.21

package/lib/agent.ts

@@ -75,6 +75,8 @@ const TaskUpData = z.object({
   worktreePath: z.string(),
   codeServerPort: z.number().int().positive().optional(),
   previewPorts: PreviewPortRuntimesSchema.optional(),
+  /** Degraded start: uai-init failed twice; surfaced as a feed system note. */
+  initWarning: z.string().optional(),
 });
 export type TaskUpResult = z.infer<typeof TaskUpData>;
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@runuai/host",
-  "version": "0.8.19",
+  "version": "0.8.21",
   "description": "Uai host — runs ephemeral AI coding tasks in Docker on a machine you control.",
   "license": "MIT",
   "author": "Diogo Perillo <diogo.perillo@gmail.com>",

package/scripts/agent/task-down.sh

@@ -20,17 +20,23 @@ UAI_TASK_ID="$task_id"
 setup_err_trap
 
 step "TASK_NOT_FOUND" "read task row"
-task_json=$(db_get_task "$task_id")
-# Idempotent: it's fine if the task row is gone — emit success and return.
-if [ "$(jq 'length' <<<"$task_json")" -lt 1 ]; then
-  emit_ok '{"alreadyGone":true}'
-  exit 0
+# Teardown must never depend on bookkeeping: a broken/missing command
+# snapshot (seen live 2026-07-21 cleaning up after a failed task-up —
+# db_get_task errored, the trap reported TASK_NOT_FOUND, and the container
+# survived) still tears the stack down by convention-derived names.
+task_json=$(db_get_task "$task_id" 2>/dev/null) || task_json="[]"
+row_count=$(jq 'length' <<<"$task_json" 2>/dev/null) || row_count=0
+if [ "$row_count" -lt 1 ]; then
+  log "task row unavailable — tearing down by derived names"
+  worktree_path=""
+  compose_project="$(compose_project_for_task "$task_id")"
+  projects_json="[]"
+else
+  worktree_path=$(jq -r '.[0].worktree_path // ""' <<<"$task_json")
+  compose_project=$(jq -r '.[0].compose_project // ""' <<<"$task_json")
+  projects_json=$(db_get_projects_for_task "$task_id" 2>/dev/null) \
+    || projects_json="[]"
 fi
-
-worktree_path=$(jq -r '.[0].worktree_path // ""' <<<"$task_json")
-compose_project=$(jq -r '.[0].compose_project // ""' <<<"$task_json")
-
-projects_json=$(db_get_projects_for_task "$task_id")
 projects_root="$UAI_WORKSPACE_ROOT/projects"
 
 # task_dir falls back to the derived path when the row's worktree_path is

package/scripts/agent/task-up.sh

@@ -198,7 +198,15 @@ while IFS= read -r project_obj; do
   [ -n "$project_default" ] || project_default="main"
 
   mkdir -p "$(dirname "$worktree_target")"
-  if git -C "$mirror_dir" rev-parse --verify --quiet \
+  if [ -e "$worktree_target/.git" ]; then
+    # Resume (ADR-028): the task worktree survived the stop — reuse it as-is
+    # (a linked worktree's .git is a FILE, hence -e). Re-running `git
+    # worktree add` would refuse ("already exists" / branch exists) and
+    # errored EVERY resume of a worktree-preserving stop within seconds
+    # (found live 2026-07-21). The task branch and any uncommitted work are
+    # exactly what the user expects back.
+    log "worktree for $project_slug already present — resuming with it"
+  elif git -C "$mirror_dir" rev-parse --verify --quiet \
        "refs/remotes/origin/$project_default" >/dev/null 2>&1; then
     # Normal case: branch the task worktree off origin/<defaultBranch>.
     step "WORKTREE_FAILED" "git worktree add ($project_id)"
@@ -537,7 +545,21 @@ if [ -f "$uai_identity_key" ]; then
     >/dev/null 2>&1 || true
 fi
 
-docker exec "$app_container" /usr/local/bin/uai-init >/dev/null
+# uai-init (workspace deps + code-server) is re-runnable, and its death must
+# not error a task whose container is already up (graceful degradation): a
+# SIGKILLed exec session (observed live 2026-07-21, exit 137 with no timeout
+# anywhere on this path) used to fail the whole task over a step anyone can
+# re-run. Retry once, then come up DEGRADED with a warning the host surfaces
+# as a system note in the task feed.
+init_warning=""
+if ! docker exec "$app_container" /usr/local/bin/uai-init >/dev/null; then
+  log "warning: uai-init failed — retrying once"
+  sleep 2
+  if ! docker exec "$app_container" /usr/local/bin/uai-init >/dev/null; then
+    init_warning="Workspace init (deps + editor) failed twice — the task is up, but the editor and installed dependencies may be missing. Retry from a task terminal with: /usr/local/bin/uai-init"
+    log "warning: $init_warning"
+  fi
+fi
 
 # -----------------------------------------------------------------------------
 # 7. Discover code-server + preview ports, mark running, unlock.
@@ -580,7 +602,9 @@ emit_ok "$(jq -nc \
   --arg cp "$compose_project" \
   --arg wp "$task_dir" \
   --arg csport "${code_server_port:-}" \
+  --arg iw "${init_warning}" \
   --argjson previewPorts "$preview_ports_runtime_json" \
   '{composeProject:$cp,worktreePath:$wp}
    + (if $csport == "" then {} else {codeServerPort:($csport|tonumber)} end)
+   + (if $iw == "" then {} else {initWarning:$iw} end)
    + {previewPorts:$previewPorts}')"

package/src/index.ts

@@ -123,6 +123,14 @@ export const hostCommands: HostCommands = {
     if (result.ok) {
       recordTaskUpResult(input.task.id, result.value);
       recordHostEvent(input.task.id, "task.started");
+      // Degraded start (uai-init failed twice): the task is up, but say so
+      // in the feed — a silent half-start reads as a broken product.
+      if (result.value.initWarning) {
+        getOrchestrator().emitSystemNote(
+          input.task.id,
+          result.value.initWarning,
+        );
+      }
       // GitHub auth for the container is best-effort (ADR-027) and runs in the
       // background — it must never block or fail task-up. Awaiting it here would
       // couple the command result to a network token-exchange: a slow/hung

package/src/protocol.ts

@@ -93,6 +93,8 @@ export interface TaskUpResult {
   worktreePath: string;
   codeServerPort?: number;
   previewPorts?: Array<{ name: string; hostPort: number }>;
+  /** Degraded start: uai-init failed twice; the task is up without deps/editor. */
+  initWarning?: string;
 }
 
 export interface TaskDownResult {