cicd-pipeline-automator
Designs and implements AWS-native CI/CD pipelines end-to-end using CodePipeline, CodeBuild, CodeDeploy, CloudFormation, CDK, and Terraform. Generates working pipeline-as-code templates (CloudFormation AWS::CodePipeline::Pipeline, CDK Pipeline module, Terraform aws_codepipeline) wired with source (CodeCommit / GitHub / CodeStar Connection), build (buildspec.yml, VPC-aware CodeBuild), test (unit, integration, CodeGuru Reviewer security scans), and deploy (CloudFormation change-set, CodeDeploy in-place / blue-green, ECS rolling, S3, Service Catalog) stages. Handles cross-account deployment (IAM roles, KMS artifact key, resource-based policy), blue/green (Lambda traffic shifting 10/90, ECS circuit breaker, Route53 weighted), pipeline monitoring (CloudWatch Events on state changes, SNS failure notifications, manual approval), and CodePipeline V2 event-driven triggers. Emits a verdict (AUTOMATED with pipeline template | MANUAL_STEP_REQUIRED with specific gap). Use when building a CI/CD pipeline, designing.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-24
- Stars
- ★ 4
- Forks
- 0
01 前言 Front Matter
- name
- cicd-pipeline-automator
- license
- Apache-2.0
- metadata
- { "tags": "codepipeline, codebuild, codedeploy, ci-cd, devtools, automate, cloudformation, cdk, terraform", "phase": "4", "author": "Jacky Chan — AWS Community Builder", "domain": "aws-cloudops", "family": "DevTools", "version": "0.1.0", "keywords": "CodePipeline, CodeBuild, CodeDeploy, CI/CD, pipeline as code, CloudFormation, CDK, Terraform, blue/green, canary, cross-account deployment, CodeStar Connection, CodeCommit, GitHub, buildspec, change-set, manual approval, pipeline monitoring, CodePipeline V2, CodeCatalyst", "task_type": "automate", "complexity": "high", "entry_point": "false", "skill_class": "capability", "when_to_use": "Designing or implementing a CI/CD pipeline using CodePipeline, CodeBuild, CodeDeploy, CloudFormation, CDK, or Terraform. Setting up source (CodeCommit, GitHub, CodeStar Connection), build (buildspec.yml, VPC-aware CodeBuild), test (unit, integration, CodeGuru), or deploy (CloudFormation change-set, CodeDeploy in-place or blue/green, ECS rolling, S3, Service Catalog) stages. Wiring cross-account deployment (IAM roles, KMS key, bucket policy). Implementing blue/green (Lambda traffic shifting, ECS circuit breaker, Route53 weighted). Adding pipeline monitoring (CloudWatch Events, SNS, manual approval). Migrating to CodePipeline V2 or troubleshooting pipeline failures.", "requires_llm": "true", "verdict_shape": "AUTOMATED | MANUAL_STEP_REQUIRED", "lifecycle_status": "active", "invocation_schema": "Input: either (a) a deployment scenario describing app, source, build, test, deploy targets, and constraints (language, environment, cross-account, blue/green), OR (b) an existing pipeline definition + the operation (create, update, troubleshoot, migrate V2). Output: deterministic OPERATION/VERDICT/REQUIREMENTS/PIPELINE_TEMPLATE/ MANUAL_GAPS block per operation, where VERDICT is AUTOMATED (full template generated) or MANUAL_STEP_REQUIRED (specific gap blocks automation).", "supports_pipeline": "true", "activation_triggers": "build CI/CD pipeline, create CodePipeline, design deployment pipeline, blue/green deployment, Lambda canary deploy, ECS rolling deploy, cross-account deployment, CodeStar Connection, buildspec.yml, pipeline as code, CloudFormation deploy pipeline, CDK pipelines, Terraform aws_codepipeline, CodePipeline V2 migration, pipeline not triggering, CodeBuild timeout, manual approval gate, SNS pipeline failure notification" }
- description
- Designs and implements AWS-native CI/CD pipelines end-to-end using CodePipeline, CodeBuild, CodeDeploy, CloudFormation, CDK, and Terraform. Generates working pipeline-as-code templates (CloudFormation AWS::CodePipeline::Pipeline, CDK Pipeline module, Terraform aws_codepipeline) wired with source (CodeCommit / GitHub / CodeStar Connection), build (buildspec.yml, VPC-aware CodeBuild), test (unit, integration, CodeGuru Reviewer security scans), and deploy (CloudFormation change-set, CodeDeploy in-place / blue-green, ECS rolling, S3, Service Catalog) stages. Handles cross-account deployment (IAM roles, KMS artifact key, resource-based policy), blue/green (Lambda traffic shifting 10/90, ECS circuit breaker, Route53 weighted), pipeline monitoring (CloudWatch Events on state changes, SNS failure notifications, manual approval), and CodePipeline V2 event-driven triggers. Emits a verdict (AUTOMATED with pipeline template | MANUAL_STEP_REQUIRED with specific gap). Use when building a CI/CD pipeline, designing.
- compatibility
- Agent runtime that reads SKILL.md (Claude Code, Cursor, Windsurf, Codex, Gemini). No AWS CLI required for offline plan classification. Live-account operations use aws codepipeline create-pipeline, update-pipeline, get-pipeline, get-pipeline-state, start-pipeline- execution, list-pipelines, aws codebuild create-project, batch-get- projects, aws codedeploy create-application, aws cloudformation create-change-set, execute-change-set, aws codestar-connections create-connection (AWS CLI v2, SSO...
02 资源文件 20
- 参考文档eval/test-cases.yaml9.5 KB
- 参考文档evals/baselines/cdk-pipeline-with-blue-green-lambda-automated.md867 B
- 参考文档evals/baselines/cross-account-kms-gap-manual.md745 B
- 参考文档evals/baselines/ecs-rolling-circuit-breaker-automated.md1.3 KB
- 参考文档evals/baselines/greenfield-cfn-pipeline-automated.md1.6 KB
- 参考文档evals/baselines/v1-pipeline-migration-manual.md756 B
- 参考文档evals/eval_config.yaml2.0 KB
- 参考文档evals/evals.json4.4 KB
- 参考文档evals/prompts/cdk-pipeline-with-blue-green-lambda-automated.md1.4 KB
- 参考文档evals/prompts/cross-account-kms-gap-manual.md1.4 KB
- 参考文档evals/prompts/ecs-rolling-circuit-breaker-automated.md1.2 KB
- 参考文档evals/prompts/greenfield-cfn-pipeline-automated.md1.1 KB
- 参考文档evals/prompts/v1-pipeline-migration-manual.md931 B
- 参考文档examples/README.md13.0 KB
- 参考文档references/advanced-patterns.md11.1 KB
- 参考文档references/diagnostic-commands.md4.1 KB
- 参考文档references/error-handling.md4.4 KB
- 参考文档references/pipeline-templates-and-deployment-strategies.md23.0 KB
- 参考文档references/worked-examples.md8.4 KB
- 参考文档skill.usf.yaml4.6 KB
03 SKILL.md 正文
CI/CD Pipeline Automator
What this skill does
Capability overview moved verbatim to references/advanced-patterns.md. Load on demand for the full capability summary (the Quick navigation table covers the essentials).
Quick navigation
| Section | What it covers | When to read it |
|---|---|---|
| § Quick reference | Verdict thresholds + requirement checklist + stage matrix | Before any build |
| § Mindset | Why pipeline-as-code, the source-credential trap, cross-account KMS | Understanding the design model |
| § Pre-flight | Requirement gate — source, build env, deploy target, IAM, KMS | Before emitting any template |
| § Process | Per-stage design: source, build, test, deploy, monitoring | When designing each stage |
| § Patterns | Pipeline templates: CloudFormation, CDK, Terraform | Boilerplate lookup |
| § Output format | Structured output template with VERDICT, PIPELINE_TEMPLATE, MANUAL_GAPS | Formatting the response |
| § Anti-Patterns | NEVER list — common mistakes that break pipelines silently | Review before emitting |
| § Pre-flight safety | Additional checks before any create/update CLI | Defense-in-depth |
Quick reference — verdict thresholds
| Verdict | Trigger condition | Action |
|---|---|---|
AUTOMATED |
All requirements satisfied (source cred, build env, deploy role, KMS key, artifact bucket, action IAM policies). Template generated. | Emit complete pipeline-as-code template, ready to apply |
MANUAL_STEP_REQUIRED |
One or more requirements missing (no CodeStar Connection, no cross-account role, no KMS key, environment mismatch). | Emit the specific gap and the exact CLI / IaC snippet to close it. Hold the partial template as a draft. |
Requirement checklist (all must be satisfied for AUTOMATED):
- Source provider credential — CodeCommit (no extra cred), GitHub
via CodeStar Connection (connection ARN in
PendingorAvailablestate), S3 (bucket exists with versioning). - Build environment — CodeBuild image supports the language
runtime (check
aws codebuild list-curated-environment-images). - Build VPC (if private resources) — VPC config with private subnets + security groups for CodeBuild if it needs to reach private VPC resources (RDS, internal ALB).
- Deploy role(s) — IAM role with trust policy for the deploy
provider (
cloudformation.amazonaws.com,codedeploy.amazonaws.com,ecs-tasks.amazonaws.com) and the deploy-action permissions. - KMS key — customer-managed CMK for artifact encryption, with
key policy granting the pipeline role + cross-account deploy roles
kms:Encrypt/Decrypt/GenerateDataKey. - Artifact bucket — S3 bucket with versioning, bucket policy granting the pipeline role, KMS-compatible SSE config.
- Cross-account (if applicable) — deploy role in target account
with
Principal: codepipeline.amazonaws.com+ source-account condition; artifact bucket policy grants target account. - Manual approval (if requested) — SNS topic for approval notifications + NotificationArn in the approval action config.
- Monitoring — EventBridge rule on
codepipelinesource matchingcodepipeline-pipeline-pipeline-execution-failed, target SNS topic for failure pages.
Stage matrix moved verbatim to references/advanced-patterns.md. Load on demand when choosing providers per stage (source/build/test/deploy/approval/invoke + pitfalls).
Mindset
One-line takeaway: pipeline-as-code is the only sustainable way to manage CI/CD at scale. Hand-built pipelines in the console have no review trail, no rollback, no diff. Driven by three CodePipeline realities:
The three CodePipeline realities moved verbatim to references/advanced-patterns.md. Load on demand for design rationale (source-credential trap, cross-account KMS, V2 triggers vs V1 polling).
Pre-flight: requirement gate
Run before emitting any template. Missing requirements produce MANUAL_STEP_REQUIRED with the exact gap.
Live-account pre-flight (skip if offline plan audit): Full command list moved verbatim to references/diagnostic-commands.md. Load on demand for live pre-flight (pipelines, connections, build images, deploy roles, KMS key, artifact bucket, STS).
Malformed input: rule moved verbatim to references/diagnostic-commands.md.
If the scenario is missing required fields, emit VERDICT: MANUAL_STEP_REQUIRED with GAP: Scenario missing required field <field>. Provide <field> to proceed.
| Requirement | Effect on automation |
|---|---|
CodeStar Connection in Pending state |
MANUAL_STEP_REQUIRED — operator must complete OAuth handshake (cannot be automated); template emitted in draft with connection ARN placeholder. |
| No CodeStar Connection at all | MANUAL_STEP_REQUIRED — emit aws codestar-connections create-connection snippet, then complete the handshake. |
| Build image does not support runtime | MANUAL_STEP_REQUIRED — emit alternative image or custom-image buildspec. |
| Deploy role missing | MANUAL_STEP_REQUIRED — emit the IAM role CloudFormation snippet (trust + policy). |
| KMS key missing or no cross-account grant | MANUAL_STEP_REQUIRED — emit aws kms create-key + key policy grant snippet. |
| Artifact bucket missing | MANUAL_STEP_REQUIRED — emit S3 bucket CloudFormation snippet with versioning + KMS SSE. |
| Manual approval requested without SNS topic | MANUAL_STEP_REQUIRED — emit aws sns create-topic snippet. |
| EventBridge rule missing for failure notifications | AUTOMATED (the rule is part of the emitted template). |
Process — pipeline design (apply in order)
Step 0: Expert knowledge — non-obvious CodePipeline behaviors
All Step 0 behaviors moved verbatim to references/advanced-patterns.md. Load on demand when a design depends on a non-obvious CodePipeline behavior (pipeline vs action roles, encryptionKey, triggers, cross-account, execution modes).
Step 1: Source stage design
| Source | Provider | Credential | Trigger |
|---|---|---|---|
| CodeCommit | CodeCommit (AWS) |
None (IAM-native) | EventBridge on referenceUpdated |
| GitHub | CodeStarConnection (AWS) |
CodeStar Connection ARN | V2 trigger OR EventBridge rule |
| GitHub (legacy) | GitHub (ThirdParty) |
OAuth token (DEPRECATED) | Polling or webhook (deprecated) |
| S3 | S3 (AWS) |
Pipeline role on bucket | EventBridge on ObjectCreated |
| ECR | ECR (AWS) |
Pipeline role on ECR | EventBridge on ImagePushed |
V2 trigger example moved verbatim to references/pipeline-templates-and-deployment-strategies.md. Load on demand when wiring V2 Git triggers (Triggers → GitConfiguration → Push → Branches / FilePathIncludes).
Step 2: Build stage design (CodeBuild)
Key decisions:
- Managed image vs custom image. Managed images
(
aws/codebuild/standard:7.0) cover common runtimes. Custom images (ECR-hosted) for proprietary runtimes or pre-installed dependencies. - VPC config. If the build needs to reach private VPC resources
(RDS, internal ALB, private ECR), set
VpcConfigwith private subnets + security groups. Add VPC endpoints or NAT for internet access. - Cache.
LOCALcache (source, layers, custom) orS3cache. S3 cache is shared across build hosts and is faster for large dependency trees. - Artifacts.
Type: CODEPIPELINEfor pipeline-managed artifacts. Secondary artifacts for separate build outputs. - Environment variables. Plain text or Secrets Manager / Parameter Store references. Never hardcode secrets in buildspec.
buildspec.yml minimum moved verbatim to references/pipeline-templates-and-deployment-strategies.md. Load on demand when emitting the build stage.
Step 3: Test stage design
| Test type | Implementation | Quality gate |
|---|---|---|
| Unit tests | CodeBuild build phase |
Non-zero exit code fails the build |
| Integration tests | CodeBuild post_build phase |
Non-zero exit code fails the build |
| Code coverage | CodeBuild + CodeGuru Profiler | Threshold gate in buildspec |
| CodeGuru Reviewer | CodePipeline action (separate stage) | Findings published to the PR; fail on critical |
| Security scans | Third-party action (Snyk, Checkov, Trivy) | Buildspec exit code on critical findings |
| Load tests | CodeBuild + Locust / k6 | Threshold gate on p99 latency |
Step 4: Deploy stage design
All deploy-strategy blocks (CloudFormation change-set, CodeDeploy in-place and blue/green, ECS circuit breaker, Route 53 weighted) moved verbatim to references/pipeline-templates-and-deployment-strategies.md. Load on demand when designing the Deploy stage.
Step 5: Cross-account deployment design
The three IAM pieces + trust-policy JSON + KMS/bucket policy grants moved verbatim to references/pipeline-templates-and-deployment-strategies.md. Load on demand for cross-account wiring (source-account role, target trust with aws:SourceAccount, artifact KMS decrypt).
Step 6: Pipeline monitoring design
EventBridge failure-rule and manual-approval blocks moved verbatim to references/pipeline-templates-and-deployment-strategies.md. Load on demand for monitoring/approval wiring (missing NotificationArn = silent gate).
Step 7: Emit template (AUTOMATED) or gap (MANUAL_STEP_REQUIRED)
If all pre-flight requirements pass, emit the complete pipeline-as-code template (CloudFormation / CDK / Terraform, per the operator's preference). The template includes: The includes list (CodePipeline / CodeBuild / IAM / KMS / S3 / Events / SNS resources) moved verbatim to references/worked-examples.md. Load on demand when emitting the AUTOMATED template.
If any requirement is missing, emit MANUAL_STEP_REQUIRED with the
specific gap and the exact CLI / IaC snippet to close it.
Patterns — pipeline-as-code templates
All three full templates (CloudFormation, CDK, Terraform) moved verbatim to references/pipeline-templates-and-deployment-strategies.md. Load on demand for boilerplate lookup.
Diagnostic flows
All five flows (pipeline not triggering, CodeBuild timeout, empty change-set, IAM AccessDenied, KMS cross-account) moved verbatim to references/diagnostic-commands.md. Load on demand when troubleshooting a running pipeline.
Output format (per operation)
Template block moved verbatim to references/worked-examples.md. The same literal labels are enforced below under "STRICT output contract"; the primary worked example follows below.
Worked example — AUTOMATED (greenfield pipeline)
OPERATION: create
VERDICT: AUTOMATED
TARGET: prod-checkout-pipeline
REQUIREMENTS:
- [PASS] CodeStar Connection arn:aws:codestar-connections:us-east-1:111111111111:connection/abc-123
in Available state
- [PASS] Build image aws/codebuild/standard:7.0 supports Node 18
- [PASS] Deploy role arn:aws:iam::111111111111:role/prod-deploy exists
with trust policy for cloudformation.amazonaws.com
- [PASS] Artifact KMS key arn:aws:kms:us-east-1:111111111111:key/abc exists
with policy granting the pipeline role
- [PASS] Artifact bucket prod-checkout-artifacts exists with versioning
and KMS SSE
- [PASS] Approval SNS topic arn:aws:sns:us-east-1:111111111111:pipeline-approval
exists with 2 subscriptions
PIPELINE_TEMPLATE:
# CloudFormation AWS::CodePipeline::Pipeline (V2) — see inline template below
Resources:
Pipeline:
Type: AWS::CodePipeline::Pipeline
Properties:
PipelineType: V2
# ... full template populated with Source (CodeStar Connection),
# Build (CodeBuild), Approval, Deploy (CloudFormation change-set)
# stages wired with the verified IAM roles, KMS key, S3 bucket.
MANUAL_GAPS: (none)
NOTES:
- Pipeline is V2 (event-driven). No PollForSourceChanges needed.
- Trigger: Git push to main branch via CodeStar Connection.
- Detection lag: ~30-60s from push to Source action firing.
- Manual approval gate before Deploy stage; approvers notified via SNS.
- Deploy uses CHANGE_SET_REPLACE + CHANGE_SET_EXECUTE (safe; change-set
can be inspected before execution).
- EventBridge rule emits SNS on FAILED/CANCELED state for on-call alerting.
Worked example — MANUAL_STEP_REQUIRED (cross-account KMS gap)
Full block moved verbatim to references/worked-examples.md. Load on demand when a requirement is [FAIL] (cross-account KMS gap).
STRICT output contract
The rules below are hard constraints. Violating any one produces a pipeline template that fails silently on first run or a gap report that leaves the operator stuck. Self-check EVERY emitted block against these rules before returning.
Required output structure
Every response MUST be a single block using these literal labels, in this order. Do NOT substitute markdown headings or camelCase variants.
OPERATION: <create | update | troubleshoot | migrate-v2>
VERDICT: AUTOMATED | MANUAL_STEP_REQUIRED
TARGET: <pipeline-name>
REQUIREMENTS:
- [PASS] <requirement description>
- [FAIL] <requirement description> — <gap>
PIPELINE_TEMPLATE: <inline CloudFormation / CDK / Terraform, or "(held in draft)">
MANUAL_GAPS:
- GAP: <gap description>
REMEDIATION: <exact CLI or IaC snippet to close the gap>
REASON: <why this cannot be automated>
NOTES: <trigger model, detection lag, cross-account caveats>
FORBIDDEN output patterns
-
NEVER emit
VERDICT: AUTOMATEDwhen any requirement is[FAIL]. A single[FAIL]MUST produceVERDICT: MANUAL_STEP_REQUIRED. The AUTOMATED verdict certifies that the template is apply-ready — any unresolved gap invalidates that certification. -
NEVER emit a
[FAIL]requirement without a correspondingMANUAL_GAPSentry. Every[FAIL]line in REQUIREMENTS MUST have a matching GAP/REMEDIATION/REASON block in MANUAL_GAPS with the exact CLI or IaC snippet that closes the gap. A[FAIL]with no remediation leaves the operator stuck. -
NEVER emit a pipeline using GitHub v1 (
ThirdParty/GitHub). The v1 OAuth integration is DEPRECATED and uses a persistent token. All GitHub/GitLab/Bitbucket sources MUST use CodeStar Connection (AWS/CodeStarConnection). If the operator provides a v1 source, emitMANUAL_STEP_REQUIREDwith the CodeStar Connection migration. -
NEVER emit a cross-account deploy role without the
aws:SourceAccountcondition on the trust policy. APrincipal: {"Service": "codepipeline.amazonaws.com"}without a source-account condition allows ANY pipeline in ANY account to assume the role. The REMEDIATION snippet MUST include the condition. -
NEVER emit a manual approval gate without
NotificationArn. An approval action without NotificationArn is a silent gate — approvers are never alerted and the pipeline sits inInProgressindefinitely. If no SNS topic exists, emitMANUAL_STEP_REQUIREDwith theaws sns create-topicsnippet. -
NEVER emit
VERDICT: AUTOMATEDwithout an EventBridge failure-notification rule in the PIPELINE_TEMPLATE. A pipeline that fails silently leaves stale code in production. The template MUST include theAWS::Events::RulematchingFAILEDandCANCELEDstates targeting an SNS topic. -
NEVER emit a V1 pipeline with
PollForSourceChanges: falsewithout the corresponding EventBridge trigger rule in the template. Setting false without the rule means the pipeline never auto-triggers. Either setPollForSourceChanges: true(with a warning about polling cost) or emit V2 with triggers. -
NEVER emit
PIPELINE_TEMPLATE: (held in draft)without listing which specific[FAIL]items blocked it. The MANUAL_GAPS block MUST enumerate every gap. A draft hold with no gap list forces the operator to diff the requirements manually.
Perfect example output — AUTOMATED
Every requirement is [PASS]. The template is complete. Copy this shape. Full example moved verbatim to references/worked-examples.md. The primary worked example above — "Worked example — AUTOMATED (greenfield pipeline)" — stays in this file.
Perfect example output — MANUAL_STEP_REQUIRED
Full example moved verbatim to references/worked-examples.md. Load on demand for the MANUAL_STEP_REQUIRED exemplar.
Self-check before emit:
- [ ] Any
[FAIL]in REQUIREMENTS → VERDICT is MANUAL_STEP_REQUIRED? - [ ] Every
[FAIL]has a matching GAP/REMEDIATION/REASON block? - [ ] No GitHub v1 (
ThirdParty/GitHub) in the template? - [ ] Cross-account role has
aws:SourceAccountcondition? - [ ] Approval gate has
NotificationArn? - [ ] AUTOMATED template includes EventBridge failure rule?
- [ ] No
PollForSourceChanges: falsewithout EventBridge rule (V1)?
Anti-Patterns — NEVER do these things
-
NEVER emit a pipeline using GitHub v1 (
ThirdParty/GitHub). The v1 integration uses a persistent OAuth token granting access to every repo under the authorizing user. It is DEPRECATED. Always use CodeStar Connection (AWS/CodeStarConnection). -
NEVER emit a pipeline without a customer-managed KMS key on the artifact store. The default (no
encryptionKey) writes artifacts to S3 without KMS-managed access control — anyone withs3:GetObjectcan read every artifact (source archives, build outputs, templates). -
NEVER emit a pipeline role (
pipeline.roleArn) withAdministratorAccess. The pipeline role orchestrates stages; it does not need broad service permissions. Grant least-privilege:codepipeline:StartPipelineExecution,s3:GetObjecton artifact bucket,codebuild:StartBuild, etc. -
NEVER emit the same IAM role for the pipeline service principal and the action deploy principal. The pipeline role is assumed by
codepipeline.amazonaws.com; the deploy role bycloudformation.amazonaws.com/codedeploy.amazonaws.com/ecs-tasks.amazonaws.com. Combining them widens the blast radius and creates a trust-policy conflict. -
NEVER emit a cross-account deploy role with
Principal: {"Service": "codepipeline.amazonaws.com"}without a source-accountCondition. Without the condition, ANY pipeline in ANY account can assume the role. -
NEVER emit a manual approval gate without
NotificationArn. The gate exists on paper but approvers are never alerted — the pipeline sits inInProgressindefinitely until someone manually checks the console. -
NEVER emit a V1 pipeline with
PollForSourceChanges: falsewithout also emitting the EventBridge rule for source change detection. V1 pipelines need EITHER polling OR an event rule — setting false without the rule means the pipeline never auto-triggers. -
NEVER emit a CodeBuild project in a public subnet when it needs to reach private VPC resources. CodeBuild in a public subnet cannot reach private subnets without VPC peering. Use private subnets + NAT or VPC endpoints.
-
NEVER emit a deploy stage without a quality gate. A pipeline that deploys on every commit without tests is a continuous-deployment footgun — a faulty commit causes an outage. Add unit, integration, and (optionally) security scan stages before Deploy.
-
NEVER emit
ActionMode: CREATE_UPDATEfor CloudFormation in a production pipeline. UseCHANGE_SET_REPLACE+CHANGE_SET_EXECUTE— the change-set can be inspected before execution, and an empty change-set can be handled gracefully. -
NEVER emit a CodeDeploy blue/green for Lambda without PreTrafficHook / PostTrafficHook. The hooks are the only programmatic way to fail the deployment if smoke tests fail during the canary window. Without hooks, a broken version ships to 100% after the canary interval elapses.
-
NEVER emit a V1 pipeline when V2 is available. V2 pipelines use triggers (event-driven), support pipeline-level variables, and have stage-level execution modes (QUEUED, SUPERSEDED). V1 defaults to polling.
-
NEVER auto-execute
create-pipelineorupdate-pipelinewithout the CONFIRM gate.UpdatePipelinereplaces the entire definition atomically with no diff and no rollback — always snapshot viaget-pipeline --output jsonfirst. -
NEVER emit a pipeline without an EventBridge rule for failure notifications. A pipeline that fails silently leaves stale code in production. Add a rule matching
FAILEDandCANCELEDstates targeting an SNS topic. -
NEVER emit a cross-account pipeline without verifying the artifact bucket policy grants the target account
s3:GetObject. The deploy role assumes the target role, which needs to read the artifact — without the bucket policy grant, the deploy fails withAccessDeniedon the S3 GetObject call. -
NEVER emit a V2 pipeline trigger with a wildcard branch filter (
*). This triggers on every branch push, including feature branches. Scope triggers tomain,release/*, or specific branch patterns. -
NEVER recommend deleting a pipeline as remediation without verifying no workloads depend on it.
DeletePipelineis irreversible.
Pre-flight safety checks (run before any remediation CLI)
-
MANDATORY CONFIRMATION GATE. Before any state-changing operation (
create-pipeline,update-pipeline,delete-pipeline,start-pipeline-execution,disable-stage-transition,enable-stage-transition), emit:CONFIRM: About to <operation> on pipeline <name> in account <account> region <region>. This affects <consequence>. Proceed? (yes/no). Do NOT execute until the operator confirms. -
Snapshot the pipeline before update.
aws codepipeline get-pipeline --name <name> --output json > /tmp/<name>-backup-$(date +%s).json.UpdatePipelinereplaces the entire definition atomically with no diff and no rollback. -
Verify the pipeline exists before update.
aws codepipeline get-pipeline --name <name>— fail closed if it returns an error. -
Before emitting a cross-account deploy role, verify the target account ID and the trust policy condition. A misconfigured condition can allow any account to assume the role.
-
Before emitting a CodeStar Connection, verify the connection state. A
Pendingconnection requires the operator to complete the OAuth handshake in the console — this cannot be automated. Emit the connection ARN as a placeholder. -
Before emitting a V2 migration, verify the pipeline is eligible. V1 pipelines using deprecated features (e.g., GitHub v1 OAuth) must migrate the source first, then upgrade to V2.
-
Prefer additive changes (add a stage, add an approval gate) over destructive changes (remove a stage, delete a pipeline) — additive changes are reversible and do not risk breaking existing deployment workflows.
Recent AWS features (2024-2026)
Moved verbatim to references/advanced-patterns.md. Load on demand for 2024-2026 CodePipeline / CodeBuild / CodeDeploy / CodeCatalyst feature coverage.
Error handling — procedure-level pipeline failures
All five failure branches (CodeBuild VPC errors, empty change-set, cross-account AssumeRole, CodeConnections revision, approval timeout) moved verbatim to references/error-handling.md. Load on demand when a pipeline-build or deploy step fails.
Edge cases
All three edge cases (approval auto-reject timeout, cross-partition source, DeletionPolicy: Retain) moved verbatim to references/advanced-patterns.md. Load on demand for these rare cases.
References (load on demand)
- references/pipeline-templates-and-deployment-strategies.md — full CloudFormation/CDK/Terraform templates; deploy strategies (Step 4); cross-account IAM (Step 5); monitoring/approval wiring (Step 6); buildspec and V2 triggers.
- references/worked-examples.md — output-format template, MANUAL_STEP_REQUIRED example, perfect AUTOMATED/MANUAL exemplars, template includes list.
- references/diagnostic-commands.md — live pre-flight commands, malformed-input rule, troubleshooting flows.
- references/error-handling.md — procedure-level pipeline failure branches.
- references/advanced-patterns.md — capability overview, stage matrix, design rationale, Step 0 expert knowledge, edge cases, 2024-2026 features.
Domain
AWS CloudOps / CodePipeline CI/CD Automation & Pipeline Engineering.
AWS documentation
- AWS CodePipeline User Guide — https://docs.aws.amazon.com/codepipeline/latest/userguide/welcome.html
- CodePipeline V2 type — https://docs.aws.amazon.com/codepipeline/latest/userguide/pipeline-types.html
- CodePipeline API Reference — https://docs.aws.amazon.com/codepipeline/latest/APIReference/
- AWS CodeBuild User Guide — https://docs.aws.amazon.com/codebuild/latest/userguide/welcome.html
- CodeBuild buildspec reference — https://docs.aws.amazon.com/codebuild/latest/userguide/build-spec-ref.html
- AWS CodeDeploy User Guide — https://docs.aws.amazon.com/codedeploy/latest/userguide/welcome.html
- CodeDeploy blue/green for Lambda — https://docs.aws.amazon.com/codedeploy/latest/userguide/deployment-steps.html
- CDK Pipelines module — https://docs.aws.amazon.com/cdk/api/v2/docs/aws-cdk-lib.pipelines-readme.html
- CodeStar Connections — https://docs.aws.amazon.com/dtconsole/latest/userguide/welcome.html
- CodeCatalyst — https://docs.aws.amazon.com/codecatalyst/latest/userguide/welcome.html
- AWS CLI CodePipeline reference — https://docs.aws.amazon.com/cli/latest/reference/codepipeline/
04 同仓库更多 Skills
xray-tracing-deployer
Deploys AWS X-Ray distributed tracing with production-grade configuration: X-Ray daemon deployment (EC2 systemd, ECS sidecar, EKS DaemonSet, Lambda built-in), SDK instrumentation (Java, Python, Node.js, Go, .NET) with annotations and metadata, sampling rules (default reservoir + rate vs custom rules with URL / method / service-name predicates), service map generation, trace query and retrieval, groups and insights, CloudWatch ServiceLens integration, IAM permissions (PutTraceSegments, GetSamplingRules), and latest features (Lambda Powertools tracing, OpenTelemetry / ADOT Collector support, W3C trace context). Emits a READY_TO_DEPLOY checklist with every configuration item verified. Use when enabling X-Ray on a new application, adding tracing to ECS / EKS / Lambda, validating a sampling configuration, or generating daemon deployment CLI and IaC templates. Triggers: X-Ray, distributed tracing, X-Ray daemon, sampling rules, service map, X-Ray SDK, OpenTelemetry, ADOT, Lambda Powertools, CloudWatch ServiceLens.
wellarchitected-workload-auditor
Audits AWS Well-Architected Tool workloads for review staleness, per-pillar high-risk issue counts, milestone tracking gaps, and remediation plan completeness. Emits a deterministic verdict (STALE_REVIEW | HIGH_RISK | CONFIG_GAP | OK) per workload with per-pillar risk breakdown, milestone trend analysis, and specific remediation. Use when reviewing Well-Architected workloads, checking WA review freshness, auditing high-risk issues per pillar, validating milestone coverage, or tracking remediation plans.
wellarchitected-review-operator
Operates AWS Well-Architected Tool reviews end-to-end — creates workloads, runs pillar reviews (operational excellence, security, reliability, performance, cost optimization, sustainability, Prosperity), answers questions with risk-tier rationale, generates reports and improvement plans, creates milestones, and integrates with Trusted Advisor and Well-Architected Labs. Runs deterministic pre-checks (workload existence, lens availability, pillar coverage), emits the exact update-answer / get-consolidated-report CLI behind a CONFIRM gate, and verifies risk-tier transitions. Emits READY | BLOCKED | COMPLETED. Use when creating a workload, answering pillar questions, generating an improvement plan, creating a milestone, or integrating with Trusted Advisor. Triggers: well-architected tool, well architected review, workload review, pillar, operational excellence, security, reliability, performance, cost optimization, sustainability, prosperity, improvement plan, milestone, trusted advisor.
wafv2-web-acl-deployer
Provisions AWS WAFv2 Web ACLs with correct production defaults: scope selection (CLOUDFRONT vs REGIONAL), managed rule groups (AWSManagedRulesCommonRuleSet, SQLiRuleSet, LinuxRuleSet, WindowsRuleSet, IPReputationList, BotControlRuleSet, AmazonIpReputationList, ATP), custom rules (byte match, IP set, regex pattern, geo match, size constraint, rate-based), rule priority ordering, logging destinations (Kinesis Firehose, CloudWatch Logs, S3), resource association (ALB, API Gateway, CloudFront), rate-based rules with aggregate keys (IP, forwarded IP), Challenge and CAPTCHA actions, and account takeover prevention. Emits a READY_TO_DEPLOY checklist. Use when creating a Web ACL, attaching managed rules, configuring rate limiting, enabling WAF logging, associating a Web ACL with an ALB/API Gateway/CloudFront, or configuring CAPTCHA/Challenge actions. Triggers: create WAF Web ACL, WAFv2, managed rule groups, WAF rate limiting, WAF logging, WAF CAPTCHA, WAF Challenge action, bot control, ATP managed rules.
wafv2-web-acl-auditor
Audits AWS WAFv2 Web ACL configurations to determine whether the ACL provides effective protection: default-action posture (Allow vs Block), managed-rule-group coverage gaps, rule effectiveness (BLOCK vs COUNT, shadow/bypass rules, stale exclusions), rate-based rule correctness, logging and visibility configuration, and text-transformation bypass vectors. Emits a deterministic verdict per Web ACL. Use when reviewing a WAFv2 Web ACL for security posture, checking managed-rule coverage, validating rate-limit configuration, auditing WAF logging, or hardening a WAF before production deployment. Triggers: WAF, WAFv2, Web ACL, managed rule group, AWSManagedRules, rate-based rule, CAPTCHA, Challenge, OverrideAction, Count mode, forwarded IP, Bot Control, ATP, text transformation, WAF logging, WAF visibility, OWASP, SQLi, XSS, RFI.
waf-rule-deployer
Provisions AWS WAFv2 rule sets with production defaults: Web ACL creation (CloudFront scope in us-east-1 vs Regional scope), rule groups (managed vs custom) with capacity planning in WCU (Web ACL Capacity Units, 1500 default per ACL), managed rule groups (AWSManagedRulesCommonRuleSet, AWSManagedRulesSQLiRuleSet, AWSManagedRulesAmazonIpReputationList, AWSManagedRulesLinuxRuleSet, AWSManagedRulesWindowsRuleSet, AWSManagedRulesUnixRuleSet), custom rules (byte-match, regex pattern set, geo-match, rate-based), IP set and regex pattern set resources, rule priority ordering (lowest number evaluated first), action types (allow, block, count, CAPTCHA, Challenge), bot control managed rule group, account takeover. Triggers: create WAF rule, WAF managed rule group, WAF custom rule, WAF byte match, WAF geo match, WAF rate based rule, WAF IP set, WAF regex pattern set, WAF WCU budget, WAF CloudFront ACL, WAF logging Kinesis Firehose, WAF bot control, WAF ATP, WAF label matching, WAF Challenge action, WAF CAPTCHA.