Allow sst refresh to fail without blocking deploy

[willwashburn]

Summary

• Adds continue-on-error: true to the refresh step
• sst refresh exits non-zero when it detects state changes (even when it succeeds), which was blocking the deploy step

Test plan

• Trigger deploy and verify it proceeds past refresh to deploy

🤖 Generated with Claude Code

---

[devin-ai-integration]

Devin Review found 1 potential issue.

View 1 additional finding in Devin Review.

[devin-ai-integration]

🔴 sst deploy proceeds unconditionally after sst refresh failure, risking production deployment on stale/corrupt state

Adding continue-on-error: true to the "Refresh SST state" step means that if sst refresh --stage production fails (e.g., due to state corruption, resource drift, or permission errors), the workflow silently continues to sst deploy --stage production at line 74. Since sst refresh synchronizes the state file with actual cloud infrastructure, deploying against a stale or inconsistent state can cause resource conflicts, duplicate resources, or failed deployments that are harder to recover from. The deploy step has no if: condition checking the refresh outcome — it runs regardless. At minimum, the refresh step should have an id and the deploy step should either gate on the refresh outcome or log a prominent warning.

Prompt for agents

In .github/workflows/deploy-web.yml, instead of unconditionally swallowing the sst refresh failure, give the refresh step an id and add a conditional warning + gating logic:

1. On line 61, add `id: refresh` to the "Refresh SST state" step (keep continue-on-error: true).
2. Before the "Deploy SST app" step (line 69), add a new step that checks the refresh outcome and emits a warning:
   - name: Warn on refresh failure
     if: steps.refresh.outcome == 'failure'
     run: echo '::warning::SST refresh failed — deploying with potentially stale state'
3. Alternatively, if a refresh failure should block deployment in most cases, remove continue-on-error: true and instead add retry logic (e.g., using nick-fields/retry action) to handle transient failures while still failing on genuine state corruption.

---

Was this helpful? React with 👍 or 👎 to provide feedback.