chore(ci): route spartan/spartan-v5 merge-train notifications to #team-alpha-ci

[AztecBot]

What

Two related changes for the spartan / spartan-v5 merge-train Slack notifications:

1. Route them to #team-alpha-ci instead of #team-alpha, so the team channel stops receiving CI noise.
2. Stop silently dropping failed Slack posts in ci3/slack_notify, so a mis-delivered notification (e.g. the bot isn't a member of the target channel) is visible in the CI log instead of vanishing.

(1) Notification sites re-routed

File	Notification	Before → After
ci3/merge_train_failure_slack_notify	"PR merged into …", CI failure, and dequeue notifications for spartan / spartan-v5 (each also kicks off a ClaudeBox investigation)	#team-alpha → #team-alpha-ci
.github/workflows/merge-train-stale-check.yml	daily "merge-train/spartan… has not merged into next/v5-next" stale alert	#team-alpha → #team-alpha-ci
.github/workflows/network-healthcheck.yml	"Starting network healthcheck …" kickoff + the ClaudeBox respond_to_user summary	#team-alpha → #team-alpha-ci
scripts/socket-fix-ci.sh	Socket dependency-vulnerability fix notification	#team-alpha → #team-alpha-ci
ci3/merge_train_stale_check	usage-example comment only	updated for consistency

ci3/run_test_cmd — flake notifications for spartan / spartan-v5 already post to #team-alpha-ci (since #23219, 2026-05-12).

(2) Surface Slack post failures (ci3/slack_notify)

Previously the chat.postMessage curl ended in &>/dev/null, so any failure — most commonly not_in_channel when the bot isn't a member of the target channel — disappeared with no trace. This is the likely reason flake notifications to #team-alpha-ci were never observed despite the routing being correct: the posts were failing silently. The script now captures the API response and logs a clear error to stderr (slack_notify: failed to post to Slack channel '<chan>': <error>) on any non-ok response or curl/network failure. It stays best-effort — the failure is logged, the caller is not failed.

Base & backport

Based on merge-train/spartan (targets next), and labeled backport-to-v5-next so the same change is carried over to the v5-next line. The network-healthcheck and merge-train-stale-check schedules ultimately run from the default branch once this reaches next via the train; the per-train CI scripts (merge_train_failure_slack_notify, slack_notify, socket-fix-ci.sh) run from the checked-out train head.

Left untouched (intentionally)

• .claude/skills/merge-trains/SKILL.md team-mapping table, ci3/dashboard/.../ci-health-report.html, spartan/testnet-runbook.md — these reference the owning team (#team-alpha) or are a manual human runbook step, not automated CI notifications.
• Other merge-train channel mappings (#team-bonobos, #honk-team, #help-ci, #dev-rels, #team-fairies) are unrelated and unchanged.

Notes

• No new Slack scopes are required — every call uses the same SLACK_BOT_TOKEN + chat.postMessage. The bot must be a member of #team-alpha-ci for delivery; once (2) lands, a missing membership shows up as a not_in_channel line in the CI log. The ClaudeBox app must also be a member so the healthcheck summary and failure/dequeue kickoff replies can post.

---

Created by claudebox · group: slackbot · requested by Santiago Palladino · Slack thread

[AztecBot]

✅ Successfully backported to backport-to-v5-next-staging #24971.