Ipa Deploy

/ipa-deploy — Deploy Infrastructure Pattern

This skill executes the deployment of a composed infrastructure pattern by running generated Makefiles. It validates prerequisites, displays a deployment plan, executes the build and deploy phases, verifies stack states, diagnoses failures, and reports results.

Lifecycle: /ipa-init → /ipa-compose → /ipa-prepare → /ipa-deploy

What This Skill Does

1. Validates prerequisites (.env, Makefiles, AWS credentials, required tools)
2. Displays a deployment plan via make -n (Make dry-run)
3. Runs make -f scripts/build.mk build (always — Make handles no-op)
4. Runs make -f scripts/deploy.mk deploy (deploys all stacks in dependency order)
5. Verifies deployed stacks reach *_COMPLETE status 5a. Runs make -f scripts/post-deploy.mk post-deploy (config generation, S3 upload, CDN invalidation, auth wiring)
6. Reports results with stack outputs, endpoints, and next steps

What This Skill Does NOT Do

• Does not read pattern files, stack SKILL.md, or TROUBLESHOOT.md — Makefiles are the only deployment contract
• Does not generate Makefiles — that's /ipa-compose's job
• Does not create IAM roles — that's /ipa-security's job
• Does not tear down infrastructure — use /ipa-destroy
• Does not generate CloudFormation templates — templates are either static (infra/cfn/) or generated
• Does not call AWS APIs directly — delegates to make which calls aws cloudformation CLI commands
• Does not support per-stack targeting — always deploys the full pattern via aggregate targets
• Does not modify .env — it is a read-only consumer of configuration

Information Sources

Step 1: Pre-Flight Validation

Run all checks below and collect results. Report all failures at once with per-item remediation — do not stop at the first failure.

AWS Credential Resolution

All AWS CLI commands in this skill must be run with .env sourced into the shell first. This sets AWS_PROFILE and AWS_REGION as environment variables, which the AWS CLI reads automatically via its default credential chain. Do NOT pass --profile or --region flags explicitly.

Pattern for every AWS CLI invocation:

source .env 2>/dev/null; aws <command>

In CI/CD (where .env may not exist), the 2>/dev/null silently skips, and the default credential chain (IAM role) takes over.

1.1 Verify .env Exists

Check that .env exists at the project root and is non-empty.

If missing: ".env not found. Run /ipa-init to initialize project configuration."

1.2 Verify Required .env Variables

Read .env and verify these variables exist and are non-empty:

Store all variable values for use in subsequent steps.

1.3 Verify Deployment Scripts Exist

Check that scripts/deploy.mk exists and contains a deploy target.

If missing: "scripts/deploy.mk not found. Run /ipa-compose to generate deployment artifacts."

1.3a Verify Prepare Stacks Are Deployed

Check that scripts/prepare.mk exists (it is always generated by /ipa-compose).

1. Parse the prepare: aggregate target to extract prepare stack names.

2. If the prepare: target is a no-op (@echo "No prepare targets..."), skip — no prepare stacks to check.

3. For each prepare stack, check status:

   source .env 2>/dev/null; aws cloudformation describe-stacks --stack-name {stack-name} --query 'Stacks[0].StackStatus' --output text
   

4. If any prepare stack does not exist or is not in a *_COMPLETE state, STOP with:

   Prerequisite infrastructure must be deployed before deploying the solution.
   Run `/ipa-prepare` to deploy one-time prerequisites.
   
   Example: /ipa-prepare
   

   Do NOT auto-invoke /ipa-prepare.

If scripts/prepare.mk is missing: "A solution must be composed before deployment. Run /ipa-compose to generate deployment artifacts."

1.4 Verify Build Script Exists

Check that scripts/build.mk exists.

If missing: "scripts/build.mk not found. Run /ipa-compose to generate build artifacts."

1.5 Verify Security Stack Is Deployed

Run:

source .env 2>/dev/null; aws cloudformation describe-stacks --stack-name {APP_NAMESPACE}-{APP_ENV}-security --query 'Stacks[0].StackStatus' --output text

Expected: CREATE_COMPLETE or UPDATE_COMPLETE.

If not deployed: "Security stack {APP_NAMESPACE}-{APP_ENV}-security is not deployed. Run /ipa-compose to configure security and compose infrastructure."

1.6 Verify AWS Credentials

Run:

source .env 2>/dev/null; aws sts get-caller-identity

If fails: "AWS credentials are invalid or expired for profile {AWS_PROFILE}. Refresh your credentials and try again."

1.7 Verify Make Is Installed

Run which make.

If missing: "GNU Make is not installed. On macOS it is pre-installed. On Linux: sudo apt install make."

Validation Summary

If any checks fail, display all failures in a single summary table and STOP:

Pre-flight validation failed:

  ✗ .env variable APP_NAMESPACE missing — Run /ipa-init
  ✗ AWS credentials expired — Refresh credentials for profile {AWS_PROFILE}

Fix the above issues and re-run /ipa-deploy.

If all checks pass: "Pre-flight validation passed. All prerequisites verified."

Step 2: Display Deployment Plan + Confirmation

Run Make dry-run to show exactly what will execute:

make -n -f scripts/deploy.mk deploy

Parse the output and display a human-readable deployment plan:

Deployment Plan: {APP_NAMESPACE}-{APP_ENV}

  Stack                              Action
  ─────────────────────────────────  ──────
  {APP_NAMESPACE}-{APP_ENV}-ddb-passengers  create/update
  {APP_NAMESPACE}-{APP_ENV}-fn       create/update (depends on: ecr, cognito, ddb-passengers)
  {APP_NAMESPACE}-{APP_ENV}-apigwv2  create/update (depends on: fn, cognito)

Proceed with deployment? (yes/no):

If scripts/post-deploy.mk exists and has non-no-op targets, also run:

make -n -f scripts/post-deploy.mk post-deploy

Include in the plan display:

Post-Deploy Steps:
  ─────────────────────────────────────────────
  configure-frontend    Generate config.js from stack outputs
  upload-frontend       Sync web-client/dist/ to S3
  invalidate-cf         CloudFront cache invalidation (waits ~1-2 min)
  update-cognito-callback  Update Cognito callback URL

• If confirmed: proceed to Step 3.
• If declined: "Deployment cancelled. No changes were made."

Step 3: Execute Build Phase

Run:

make -f scripts/build.mk build

Always execute this step — Make handles no-ops. If build.mk has no real targets, the aggregate target outputs "No build targets for this pattern" and exits 0.

Display Make output as it runs.

If build fails (exit code ≠ 0): Display the full Make output, propose a fix based on the error, and STOP. Do not proceed to deploy.

Step 4: Execute Deploy

Run:

make -f scripts/deploy.mk deploy

Display Make output as it runs. Each target prints its aws cloudformation command, providing natural progress indication.

If deploy succeeds (exit code = 0): proceed to Step 5.

If deploy fails (exit code ≠ 0): go to Failure Diagnosis.

Idempotency: Re-running deploy after a partial failure is always safe. The execution layer handles all state transitions:

• Non-existent stack → CreateStack
• ROLLBACK_COMPLETE → delete, then recreate
• *_COMPLETE → UpdateStack
• No changes → succeeds silently ("No updates are to be performed")

Step 5: Post-Deploy Verification

For each stack deployed by scripts/deploy.mk, run:

source .env 2>/dev/null; aws cloudformation describe-stacks --stack-name {stack-name} --query 'Stacks[0].StackStatus' --output text

Confirm all stacks report CREATE_COMPLETE or UPDATE_COMPLETE.

Then collect outputs for reporting:

source .env 2>/dev/null; aws cloudformation describe-stacks --stack-name {stack-name} --query 'Stacks[0].Outputs'

Optionally, run an overview of all managed stacks:

source .env 2>/dev/null; aws cloudformation list-stacks --stack-status-filter CREATE_COMPLETE UPDATE_COMPLETE UPDATE_ROLLBACK_COMPLETE --query 'StackSummaries[?starts_with(StackName, `{APP_NAMESPACE}-{APP_ENV}`)]'

Step 5a: Execute Post-Deploy

Check if scripts/post-deploy.mk exists (it is always generated by /ipa-compose).

1. Parse the post-deploy: aggregate target.
2. If the target is a no-op (@echo "No post-deploy..."), skip — proceed to Step 6.
3. If targets exist:
   • Display: "Running post-deploy steps..."
   • Execute:

     make -f scripts/post-deploy.mk post-deploy
     

   • Display Make output as it runs.
   • If succeeds: Include post-deploy results in Step 6 report.
   • If fails: Display the error. Advise: "Post-deploy failed. Deployed stacks are intact — no rollback needed. Fix the issue and re-run /ipa-deploy (deploy will no-op, post-deploy will retry). Or run manually: make -f scripts/post-deploy.mk post-deploy" Do NOT roll back deployed CloudFormation stacks.

If scripts/post-deploy.mk is missing: Skip silently — patterns composed before this feature will not have post-deploy.mk. This is not an error.

Step 6: Completion Report

Display a structured report:

Deployment Complete: {APP_NAMESPACE}-{APP_ENV}

  Stack                                      Status
  ─────────────────────────────────────────  ───────────────
  {APP_NAMESPACE}-{APP_ENV}-ddb-passengers   CREATE_COMPLETE
  {APP_NAMESPACE}-{APP_ENV}-fn               CREATE_COMPLETE
  {APP_NAMESPACE}-{APP_ENV}-apigwv2          CREATE_COMPLETE

  Outputs:
  ────────
  {APP_NAMESPACE}-{APP_ENV}-fn:
    FunctionArn = arn:aws:lambda:us-east-1:123456789012:function:ipatest-dev-fn

Next steps:
  • Review deployed resources in the AWS Console
  • Run /ipa-codepipeline to set up CI/CD (optional)
  • Run /ipa-destroy to tear down infrastructure when no longer needed
  • Re-run /ipa-deploy at any time — it is safe to re-run (idempotent)

If post-deploy ran successfully, add to the completion report:

  Post-Deploy:
  ────────────
  ✓ configure-frontend    — web-client/dist/config.js generated
  ✓ upload-frontend       — synced to S3
  ✓ invalidate-cf         — CloudFront cache invalidated
  ✓ update-cognito-callback — Cognito callback URL updated

  Application URL: {AppUrl from CloudFront stack output}

Next steps:
  • Open {AppUrl} to verify the deployed application
  • Run /ipa-destroy to tear down infrastructure when no longer needed
  • Re-run /ipa-deploy at any time — it is safe to re-run (idempotent)
  • Re-run post-deploy only: make -f scripts/post-deploy.mk post-deploy

Failure Diagnosis

When a Make target fails during deployment (Step 4), follow this procedure to diagnose and recover.

1. Detect the Failed Stack

Read the Make output to identify which target failed. The failed stack name follows the --stack-name flag in the aws cloudformation command that errored.

2. Read Stack Events

source .env 2>/dev/null; aws cloudformation describe-stack-events --stack-name {failed-stack-name}

This returns recent events in chronological order with resource status and status reason.

3. Classify the Error

Match event text against the error classification table in DIAGNOSIS.md. The five categories are:

4. Execute Recovery

For simple recoveries (stuck rollback, transient — marked Auto):

1. Explain the diagnosis and proposed fix to the builder.
2. Ask: "Would you like me to fix this automatically? (yes/no):"
3. If confirmed, execute the fix:
   • Stuck rollback: source .env 2>/dev/null; aws cloudformation delete-stack --stack-name {failed-stack} followed by source .env 2>/dev/null; aws cloudformation wait stack-delete-complete --stack-name {failed-stack}, then re-run make -f scripts/deploy.mk deploy.
   • Transient: Wait 30 seconds, then re-run make -f scripts/deploy.mk deploy.

For complex fixes (permission, validation, resource conflict — marked Manual):

1. Explain the diagnosis and the specific error from cfn-events.
2. Advise the builder on what to do — see DIAGNOSIS.md for per-category guidance.
3. After the builder has applied the fix, they should re-run /ipa-deploy.

For deploy-level failures not covered by CloudFormation events (Make errors, tool missing, credential issues), see TROUBLESHOOT.md.

Error Handling

No-Op Build

If scripts/build.mk has no real targets, the build aggregate target outputs "No build targets for this pattern" and exits 0. This is expected — proceed to deployment.

Dependent Stack Failure

If a stack fails but its prerequisites succeeded, Make stops at the failed target. Already-deployed prerequisite stacks remain untouched. Re-running deploy after fixing the root cause will skip completed stacks and retry only the failed one.

Credential Expiry Mid-Deployment

If credentials expire during a long-running deployment (e.g., temporary STS tokens with short TTL), the current aws cloudformation command will fail with an access denied error. The builder should refresh credentials and re-run /ipa-deploy — idempotency ensures already-deployed stacks are not affected.

Deployment Timeout

CloudFormation deployments have a 60-minute default timeout. If a stack exceeds this, cfn-events will show the timeout. Advise the builder to check the stack status in the AWS Console — the stack may still be in progress.

Network Connectivity Loss

If connectivity is lost mid-deployment, the current command will fail. Advise the builder that infrastructure may be in a transitional state. They should:

1. Check stack status: source .env 2>/dev/null; aws cloudformation list-stacks --stack-status-filter CREATE_COMPLETE UPDATE_COMPLETE CREATE_IN_PROGRESS UPDATE_IN_PROGRESS ROLLBACK_IN_PROGRESS --query 'StackSummaries[?starts_with(StackName, {APP_NAMESPACE}-{APP_ENV})]'
2. Wait for any IN_PROGRESS stacks to reach a terminal state.
3. Re-run /ipa-deploy to complete the deployment.