Deploy
HARD GATE — Do not deploy without running tests first. Run
testor your CI suite before this skill.HARD GATE — Use this skill from a CI/CD pipeline or post-merge on
main/master. Never deploy from a feature branch.HARD GATE — The deploy skill orchestrates deployment; the
smoke-testskill validates post-deploy health. Chain them:deploy → smoke-test.
Orchestrate a full build-to-deployment pipeline: build the artifact, verify it exists and is non-empty, invoke a platform deploy tool (MCP or CLI), poll until the deploy completes or times out, then run a baseline smoke test against the live URL.
Pipeline Stages
build → verify artifact → deploy → wait/retry → smoke
| Stage | Description | Failure mode |
|---|---|---|
| Build | Execute the project's build command | Non-zero exit: report build error |
| Verify | Check artifact exists and is non-empty | Missing/empty: report artifact path |
| Deploy | Invoke platform deploy tool (MCP, Vercel CLI, rsync, etc.) | Non-zero exit: report deploy error |
| Wait | Poll deploy status every 30s up to DEPLOY_TIMEOUT (default 5 min) | Timeout: report exceeded |
| Smoke | curl -sSf $DEPLOY_URL as baseline health check | Non-200: report failure |
Process
1. Detect build command
Read project manifest files in order to determine the build command:
| Manifest | Build command |
|---|---|
package.json | npm run build (or scripts.build value) |
Cargo.toml | cargo build --release |
pyproject.toml / setup.py | Depends on build backend (poetry build, pip install -e ., etc.) |
Makefile | make build or first target named build |
AGENTS.md / CLAUDE.md | Look for build: in project commands section |
If no manifest is found, prompt the user with: "No detected build command. Pass --build 'npm run build' or specify the command."
2. Build the artifact
bashnpm run build
Or the detected command from step 1. If the build fails, exit non-zero and report the build output.
3. Verify the artifact
bashARTIFACT_DIR="${ARTIFACT_DIR:-dist}" if [ ! -d "$ARTIFACT_DIR" ] || [ -z "$(ls -A "$ARTIFACT_DIR" 2>/dev/null)" ]; then echo "FAIL: build artifact not found at $ARTIFACT_DIR" exit 1 fi
Configurable via $ARTIFACT_DIR environment variable (default: dist/).
4. Deploy to platform
Platform-agnostic — supports multiple deployment targets via environment variables:
| Platform | Env var | Example |
|---|---|---|
| Vercel | VERCEL_TOKEN, VERCEL_PROJECT_ID | vercel deploy --prod --token $VERCEL_TOKEN |
| Netlify | NETLIFY_AUTH_TOKEN, NETLIFY_SITE_ID | netlify deploy --prod --auth $NETLIFY_AUTH_TOKEN --dir $ARTIFACT_DIR |
| Platform MCP | MCP tool call | mcp deploy via your platform MCP server |
| rsync/SSH | DEPLOY_SSH_USER, DEPLOY_SSH_HOST, DEPLOY_SSH_PATH | rsync -avz $ARTIFACT_DIR/ $DEPLOY_SSH_USER@$DEPLOY_SSH_HOST:$DEPLOY_SSH_PATH |
| Custom | DEPLOY_COMMAND | Run any deploy command string |
The deploy tool is selected by which environment variables are set. If none are configured:
bashecho "No deploy target configured. Set one of: VERCEL_TOKEN, NETLIFY_AUTH_TOKEN, DEPLOY_SSH_USER+DEPLOY_SSH_HOST, DEPLOY_COMMAND, or MCP deploy tool." exit 1
5. Wait and poll status
After invoking the deploy command, poll for completion:
See REFERENCE.md
Use exponential backoff for retries on transient failures:
See REFERENCE.md
6. Baseline smoke test
See REFERENCE.md
For comprehensive health-checking, chain to the smoke-test skill:
bash# After deploy success bash scripts/run-smoke.sh "$DEPLOY_URL"
7. Three-independent-facts verification (e45s15)
Before declaring deploy success, verify three independent facts — build artifact, platform accept, live/registry reachability. See REFERENCE.md.
Verify
→ verify: command -v curl >/dev/null 2>&1 && test -f skills/smoke-test/SKILL.md
