Smoke Test

HARD GATE — Do NOT run smoke-test against a URL that hasn't been deployed yet. Always run deploy first, then smoke-test.

HARD GATE — A failed smoke test means the deployment is broken. Do NOT mark a deploy as successful until all smoke checks pass.

Validate a deployed application is healthy by running HTTP checks against live URLs. Each check asserts HTTP status, optional body signal (regex), and optional response-time threshold.

Configuration

Smoke checks live in smoke-checks.yaml at the project root:

base_url: "https://example.com"
checks:
  - name: "Homepage"
    path: "/"
    expected_status: 200
    content_signal: "welcome|ok"
    max_response_time_ms: 3000

Ad-hoc single-URL mode: DEPLOY_URL=https://host bash scripts/run-smoke.sh

Process

1. Load checks

SMOKE_CHECKS_FILE="${SMOKE_CHECKS_FILE:-smoke-checks.yaml}"
BASE_URL="${DEPLOY_URL:-$BASE_URL}"
test -f "$SMOKE_CHECKS_FILE" || test -n "$BASE_URL" || { echo "ERROR: no checks file or URL"; exit 1; }

2. Run each check

bash scripts/run-smoke.sh "${DEPLOY_URL:-}" "${SMOKE_CHECKS_FILE:-smoke-checks.yaml}"

The runner performs curl requests per check, records pass/fail per assertion, and prints a summary.

3. Assert results

• Any HTTP status mismatch → FAIL
• Missing content_signal when configured → FAIL
• Response time over max_response_time_ms → FAIL
• Exit code non-zero → deployment not healthy

4. Generate report

Capture stdout from run-smoke.sh as evidence. Persist to specs/verifications/smoke-<date>.log for release-branch.

Integration with deploy skill

DEPLOY_URL="$DEPLOY_URL" bash scripts/run-smoke.sh

Verify arc

Part of ★ VERIFY ★: verify-work → validate-contracts → smoke-test → run-evals → audit-code

Verify

→ verify: test -x scripts/run-smoke.sh && grep -q 'run-smoke.sh' skills/smoke-test/SKILL.md && ! grep -q 'See \[REFERENCE.md\](REFERENCE.md)$' skills/smoke-test/SKILL.md && echo OK