test/nemoclaw-start.test.ts (1)

1677-1816: ⚡ Quick win

Assert that seed_default_workspace_templates is actually invoked in both startup paths.

The harness defines a no-op stub, but current assertions don’t verify call-through. A regression removing the call sites could pass unnoticed.

Small assertion upgrade

-        'seed_default_workspace_templates() { :; }',
+        'seed_default_workspace_templates() { echo "ORDER:seed"; }',
...
       expect(setup.result.stdout).toContain("ORDER:configure");
+      expect(setup.result.stdout).toContain("ORDER:seed");
       expect(setup.result.stdout).toContain("VALIDATE:");

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@test/nemoclaw-start.test.ts` around lines 1677 - 1816, The test defines a
no-op stub seed_default_workspace_templates() but doesn't assert it was invoked;
modify runPreGatewaySetup's generated script to make
seed_default_workspace_templates emit a recognizable marker (e.g., printf
"ORDER:seed_default_workspace_templates\n" or "SEED_CALLED") when invoked, then
update the final assertions in the "installs and validates..." test to check
setup.result.stdout contains that marker (reference
seed_default_workspace_templates and runPreGatewaySetup to locate where the stub
is emitted and where to add the new expect on setup.result.stdout).

scripts/nemoclaw-start.sh (2)

1486-1522: ⚡ Quick win

Gate seeding behind skipBootstrap=true to avoid changing bootstrap-enabled behavior.

Right now seeding runs for any pristine workspace. If a config intentionally enables bootstrap, this may pre-populate files (without BOOTSTRAP.md) and alter first-run behavior.

Suggested guard (example)

 seed_default_workspace_templates() {
   local workspace_dir="${1:-/sandbox/.openclaw/workspace}"
   local templates_dir="${2:-}"
+  local config_file="/sandbox/.openclaw/openclaw.json"
+
+  # Only backfill when bootstrap is explicitly skipped.
+  if [ -f "$config_file" ]; then
+    python3 - "$config_file" <<'PY' >/dev/null 2>&1 || return 0
+import json, sys
+cfg = json.load(open(sys.argv[1], "r", encoding="utf-8"))
+raise SystemExit(0 if cfg.get("agents", {}).get("defaults", {}).get("skipBootstrap") is True else 1)
+PY
+  fi
+
   [ -e "$workspace_dir" ] || return 0
   if [ -L "$workspace_dir" ]; then

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@scripts/nemoclaw-start.sh` around lines 1486 - 1522, The seeding function
seed_default_workspace_templates should only run when skipBootstrap=true to
avoid altering bootstrap-enabled first-run behavior; add an early guard at the
top of seed_default_workspace_templates that checks the skipBootstrap
environment/config variable (e.g. [ "${skipBootstrap:-}" != "true" ] && return
0) before any workspace checks, so the rest of the function (the symlink check,
pristine directory check, template_dir resolution, and file copy loop for
AGENTS.md SOUL.md IDENTITY.md USER.md TOOLS.md HEARTBEAT.md) only executes when
skipBootstrap is explicitly set to "true".

---

1476-1765: Run targeted E2E for this entrypoint change before merge.

This script is startup-critical and some failure modes won’t be covered by unit tests alone; run the recommended sandbox-survival-e2e, sandbox-operations-e2e, and cloud-e2e jobs for confidence.

As per coding guidelines scripts/nemoclaw-start.sh: “This file is a sandbox entrypoint script. Changes affect every sandbox boot and are invisible to unit tests … E2E test recommendation: sandbox-survival-e2e, sandbox-operations-e2e, cloud-e2e.”

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@scripts/nemoclaw-start.sh` around lines 1476 - 1765, The change touches the
sandbox entrypoint and must be verified by running the recommended end-to-end
suites before merging: run sandbox-survival-e2e, sandbox-operations-e2e, and
cloud-e2e to exercise startup/privilege paths; focus checks on
migrate_legacy_layout, seed_default_workspace_templates,
provision_agent_workspaces, and the non-root vs root branches (including gosu
sandbox launch, gateway boot, and tmp/gateway log handling) to ensure no startup
regressions or permission/ownership failures; fix any failures uncovered and
only merge once all E2E jobs pass.