- Move the .agent @container rule after the base .agent rules. A container
query adds no specificity, so the earlier position always lost and rows
kept six fixed columns, spilling Revoke buttons onto the open drawer.
- Give .drawer min-width:0 so the compose snippet scrolls inside its <pre>
instead of widening the drawer past the viewport at 390px.
Claude-Session: https://claude.ai/code/session_01GAZsp3P7Qto6JnhipxWUJQ
- icons/gen writes to the current directory so `go generate
./internal/master/web/icons` works (output verified byte-identical)
- share the registry-host check as imageref.IsRegistryHost (with test)
- agents fragment refreshes the "N machines" header count out of band
- ignore the saved machine filter at <=760px where the machine list is hidden
- aria-current on the active machine filter link, aria-pressed on theme toggle
- CLAUDE.md: icons package, SSE ping event, swarm_cluster_id
Claude-Session: https://claude.ai/code/session_01GAZsp3P7Qto6JnhipxWUJQ
Restyles the master UI after the Ledger design: a sidebar with live counts,
a machine filter, a Live box and a light/dark toggle; the overview groups
services per Swarm cluster and machine with icons, search and the
Icon / Icon + text / Text switch; the agents page gets a revoke popover and
an add-agent panel with a connect step. Jobs, Job and Service pages get the
new shell and tokens only.
- fonts and Simple Icons are served by the master (no CDN)
- the overview reads agents too, to show offline banners and cluster managers
- revoking an agent also publishes the agents event
- the SSE keepalive is a named event (ping) the Live box counts
- the master passes its build version to the UI so agents on another
version are marked old
Claude-Session: https://claude.ai/code/session_01GAZsp3P7Qto6JnhipxWUJQ
## Summary
- `POST /update` on the master (outside OIDC, bearer `UPDATER_API_KEY`; unset → 503). Matches `image` against `image_repo` on all agents, marks `available_*` on every match, starts webhook-triggered jobs for `auto` services targeting `<image_repo>:<tag>` (followed tag when no tag is given), and waits for them up to `UPDATE_TIMEOUT` (default 5m). 404 on no match, 500 if any auto job failed or timed out, otherwise 200.
- Manual/auto toggle per service on the overview and service page (`POST /services/{id}/mode`, CSRF); auto is blocked for self, `build:` and container services with the reason shown. "Update available" badge and an "Available" row.
- `FinishJob` clears `available_*` only when the successful job was created for that exact ref, after it was reported. Migration `0003_webhook.sql`, phase-3 plan doc, README / gitea-action README / CLAUDE.md / deploy example updated. `action.yml` unchanged.
Closes#8
## Behaviour changes for CI
- Mixed results now return 500 (the old 207 is gone); no match returns 404 instead of 200.
- With a `tag`, Compose services get pinned to `<image_repo>:<tag>` via `compose.override.yaml`; the tag must be a pushed tag.
- `busy` (job already active) and `refused` (not updatable) are reported but return 200.
- The reverse proxy read timeout must exceed `UPDATE_TIMEOUT`.
## Test plan
- [x] `go test ./...` and `go test -race ./internal/agent/... ./internal/master/... ./internal/e2e/`
- [x] `go test -tags integration ./internal/master/store/` against throwaway Postgres
- [x] UI verified in browser (mode toggle, disabled reason, badge via SSE, Available row)
- [ ] Manual: webhook with manual service → `marked` + badge
- [ ] Manual: webhook with auto service → waits, `succeeded`, badge cleared, override pinned
- [ ] Manual: wrong token 401, unknown image 404, offline agent 500
- [ ] Gitea action pointed at the master's `/update` in a real CI run
Reviewed-on: #22
shcizo