chore(agent-team): deploy-r720.sh handles the dashboard frontend path + verify fix #69

Merged

amoussa1229 merged 1 commit from chore/deploy-r720-frontend-path into main

2026-06-25 14:56:23 +00:00

amoussa1229 commented

2026-06-24 23:31:32 +00:00

(Migrated from github.com)

What

Extends agent-team/scripts/deploy-r720.sh (the script /sh-deploy-r720 drives) to cover the dashboard frontend path, and fixes a verify false-fail.

SYNC_WEB=1 — build agent-team/web/dist on the Mac (the box Node v18 is too old for Vite 5+), rsync --delete it to the box (clears stale content-hashed bundles), force an agent-team-status.service restart, and verify the served index.html references the freshly-built bundle (catches a stale-dist deploy). Implies RESTART_STATUS=1.
Verify fix — the post-restart journal check used to fail on any traceback|error, but the box fires the known-benign draft-pr-monitor FileNotFoundError: 'gh' every ~30s (box has no gh CLI), so every deploy false-failed. Now strips that specific traceback block before judging; is-active / NRestarts / thread-count remain the authoritative crash-loop signals.

Why

The dashboard SPA is served by a separate service (agent-team-status.service) from web/dist, which the coordinator-only deploy path didn't handle — the redesign deploy (PR #64) was done by hand. This codifies it. See feedback_r720_agent_team_deploy memory.

Testing

bash -n clean.
Mirrors the exact manual steps used to deploy PR #64 to the live box (build-on-Mac → rsync --delete web/dist → restart status service → served-bundle check).
No behavior change to the default (coordinator-only) path when SYNC_WEB/RESTART_STATUS are unset.

Note

Pushed cleanly through the pre-push security gate (no --no-verify) after registering machine-side suppressions for the repo's standing test-key/.env.example false-positives.

## What Extends `agent-team/scripts/deploy-r720.sh` (the script `/sh-deploy-r720` drives) to cover the **dashboard frontend** path, and fixes a verify false-fail. - **`SYNC_WEB=1`** — build `agent-team/web/dist` **on the Mac** (the box Node v18 is too old for Vite 5+), `rsync --delete` it to the box (clears stale content-hashed bundles), force an `agent-team-status.service` restart, and **verify the served `index.html` references the freshly-built bundle** (catches a stale-dist deploy). Implies `RESTART_STATUS=1`. - **Verify fix** — the post-restart journal check used to fail on any `traceback|error`, but the box fires the known-benign draft-pr-monitor `FileNotFoundError: 'gh'` every ~30s (box has no `gh` CLI), so every deploy false-failed. Now strips that specific traceback block before judging; `is-active` / `NRestarts` / thread-count remain the authoritative crash-loop signals. ## Why The dashboard SPA is served by a **separate** service (`agent-team-status.service`) from `web/dist`, which the coordinator-only deploy path didn't handle — the redesign deploy (PR #64) was done by hand. This codifies it. See `feedback_r720_agent_team_deploy` memory. ## Testing - `bash -n` clean. - Mirrors the exact manual steps used to deploy PR #64 to the live box (build-on-Mac → `rsync --delete web/dist` → restart status service → served-bundle check). - No behavior change to the default (coordinator-only) path when `SYNC_WEB`/`RESTART_STATUS` are unset. ## Note Pushed cleanly through the pre-push security gate (no `--no-verify`) after registering machine-side suppressions for the repo's standing test-key/`.env.example` false-positives.

This repo is archived. You cannot comment on pull requests.

No reviewers