feat: mark rate-limited automation runs as Skipped – Limit Reached

[hieptl]

Summary

Automation runs that can't start because the organization (incl. Personal Workspaces) is at its concurrent-sandbox limit are now marked with a new terminal status SKIPPED and shown in the Activity Log as "Skipped – Limit Reached" — instead of being retried and then marked FAILED.

Why

When the org is at its limit, POST /api/v1/sandboxes returns 429 with detail.error == "CONCURRENCY_LIMIT_REACHED". The cloud backend treated this like any transient rate limit: retry 5× (10–60s each), then mark the run FAILED ("Failed to get execution context"). That's slow and misleading — retrying can't free a slot, and it's not really a failure. (This mirrors what the OpenHands integration managers already do for ConcurrencyLimitError on the conversation side — APP-1031.)

What changed

Backend (openhands/automation/)

• models.py / schemas.py — add SKIPPED to AutomationRunStatus and the API RunStatus.
• exceptions.py — new ConcurrencyLimitReachedError (a plain Exception, not PermanentDispatchError, so the automation is not disabled).
• backends/cloud.py — _concurrency_limit_detail() recognizes the limit 429 (tolerating FastAPI's {"detail": {…}} nesting and a flat {…}). It's raised inside the retried request closure, so — being a non-HTTPStatusError — it bypasses the retry predicate (no backoff) while transient 429s still retry unchanged.
• dispatcher.py — catch ConcurrencyLimitReachedError around get_execution_context and mark the run SKIPPED via mark_run_terminal (no disable).
• utils/run.py — SKIPPED joins the terminal set that stamps completed_at.

Frontend (frontend/)

• types/automation.ts — add SKIPPED to the enum (the badge's exhaustive Record then requires the entry below).
• components/automations/detail/run-status-badge.tsx — map SKIPPED → AUTOMATIONS$DETAIL$SKIPPED_LIMIT_REACHED (neutral/muted style).
• i18n/translation.json — new key, all 15 locales, en = "Skipped – Limit Reached" (en‑dash, U+2013). declaration.ts and public/locales/** are gitignored and regenerated by make-i18n.

Tests

• tests/test_backends.py — _concurrency_limit_detail discriminator (nested/flat/transient/no-marker/non-JSON/non-429) + _create_sandbox raises ConcurrencyLimitReachedError and is not retried.
• tests/test_dispatcher.py — _execute_run marks the run SKIPPED (enabled, no error_detail); regression that a generic context failure still marks FAILED.
• frontend/.../run-status-badge.test.tsx — badge renders the skipped label.

Behavior

Scenario	Before	After
Org at concurrency limit	retry 5× → FAILED ("Failed to get execution context")	immediate SKIPPED → "Skipped – Limit Reached"
Transient rate-limit 429	retried	retried (unchanged)
Automation after a skip	—	stays enabled

Notes

• No DB migration — automation_runs.status is VARCHAR(20) with no constraint (same as the CANCELLED precedent).
• Dependency — requires an OpenHands deployment that emits the CONCURRENCY_LIMIT_REACHED 429 (APP-1031/conversation-limits).
• Out of scope — the frontend run-status enum/badge is pre-existingly missing CANCELLED (a CANCELLED run would render an undefined badge). Not introduced here; recommend a separate follow-up.

Testing

# backend
uv run pytest tests/test_backends.py tests/test_dispatcher.py -q   # 65 passed
uv run ruff check openhands/automation tests

# frontend
npm run typecheck                      # exit 0 (exhaustive badge Record compiles)
npm run check-translation-completeness # all 15 locales present
npx vitest run run-status-badge        # 5 passed

Verification (Activity Log)

1. With an org/workspace at its concurrent-sandbox limit, trigger an automation (cron-due or POST /v1/{id}/dispatch).
2. Confirm the run row is SKIPPED (not FAILED), with started_at/completed_at set and error_detail null.
3. Open the automation's Activity Log and confirm the badge reads "Skipped – Limit Reached".

[github-actions]

[tofarr]

🍰