[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7AC47XY4NVD1T1942VSJSXM":3,"skill-same:01M7AC47XY4NVD1T1942VSJSXM":140},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":40,"commitSha":41,"skillDirectory":42,"forksCount":43,"repositoryHtmlUrl":44,"repositoryHomepageUrl":6,"ownerHtmlUrl":45,"ownerBlogUrl":6,"ownerDescription":46,"updatedAt":47,"install":48,"resources":61},"01M7AC47XY4NVD1T1942VSJSXM","sagemaker-endpoint-auditor",null,"Audits SageMaker real-time and async endpoints for security posture and production readiness — KMS encryption-at-rest and inter-container traffic encryption, execution-role blast radius (wildcard actions, sagemaker:*, iam:PassRole), VPC configuration (internet-facing vs private), data capture and model monitoring schedule coverage, and instance count for high availability. Emits a deterministic verdict (NO_ENCRYPTION | OVERPERMISSIVE_ROLE | NO_MONITORING | PUBLIC_ENDPOINT | CONFIG_GAP | OK) per endpoint with enumerated findings and CLI remediation. Use when reviewing SageMaker endpoints before production deployment, checking endpoint encryption posture, auditing execution-role scope, validating VPC configuration, verifying model monitoring coverage, or checking instance count for high availability.","jhkchan\u002Faws-agent-skills","jhkchan","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F1083902?v=4","01M7A2SVQYP9GQTPTDXP5JKYQ4","Jacky Chan","01M7A2SVS117YWCTZABB75HZR1",4,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: sagemaker-endpoint-auditor\ndescription: Audits SageMaker real-time and async endpoints for security posture and production readiness — KMS encryption-at-rest and inter-container traffic encryption, execution-role blast radius (wildcard actions, sagemaker:*, iam:PassRole), VPC configuration (internet-facing vs private), data capture and model monitoring schedule coverage, and instance count for high availability. Emits a deterministic verdict (NO_ENCRYPTION | OVERPERMISSIVE_ROLE | NO_MONITORING | PUBLIC_ENDPOINT | CONFIG_GAP | OK) per endpoint with enumerated findings and CLI remediation. Use when reviewing SageMaker endpoints before production deployment, checking endpoint encryption posture, auditing execution-role scope, validating VPC configuration, verifying model monitoring coverage, or checking instance count for high availability.\nlicense: Apache-2.0\ncompatibility: Agent runtime that reads SKILL.md (Claude Code, Cursor, Windsurf, Codex, Gemini). No AWS CLI required for offline config-document classification. Live-account audits use aws sagemaker describe-endpoint, describe-endpoint-config, describe-model, and list-monitoring-schedules (AWS CLI v2, SSO or key-based credentials).\nmetadata:\n  domain: aws-cloudops\n  complexity: high\n  requires_llm: 'true'\n  phase: '2'\n  supports_pipeline: 'true'\n  entry_point: 'false'\n  family: AI\u002FML\n  verdict_shape: NO_ENCRYPTION | OVERPERMISSIVE_ROLE | NO_MONITORING | PUBLIC_ENDPOINT | CONFIG_GAP | OK\n  when_to_use: Reviewing a SageMaker endpoint before production deployment, checking endpoint encryption (KMS + inter-container), auditing the execution role for wildcard permissions, validating VPC configuration (private vs internet-facing), verifying model monitoring coverage (data capture + monitoring schedule), or checking instance count for high availability.\n  activation_triggers: audit this SageMaker endpoint, is my endpoint encrypted, check endpoint VPC config, execution role too broad, is model monitoring enabled, endpoint production ready, SageMaker endpoint public, inter-container encryption\n  invocation_schema: 'Input: either (a) a SageMaker endpoint configuration bundle (Model + EndpointConfig + Endpoint metadata + optional MonitoringSchedule + optional execution-role policy document), OR (b) an endpoint name\u002FARN for live-account audit. Output: deterministic ENDPOINT\u002FVERDICT\u002FREASON\u002FFINDINGS\u002F REMEDIATION block per endpoint, where VERDICT is one of NO_ENCRYPTION, OVERPERMISSIVE_ROLE, NO_MONITORING, PUBLIC_ENDPOINT, CONFIG_GAP, OK.'\n  version: 0.1.0\n  author: Jacky Chan — AWS Community Builder\n  keywords: SageMaker, endpoint, KMS encryption, inter-container encryption, execution role, VPC config, data capture, model monitor, instance count, high availability, production readiness, endpoint audit, inference pipeline, network isolation, sagemaker:InvokeEndpoint, blast radius\n  tags: sagemaker, ai-ml, security, endpoint, encryption, vpc, monitoring, audit\n---\n\n# SageMaker Endpoint Auditor\n\n## Mindset\n\n**One-line takeaway:** a SageMaker endpoint is a running model exposed behind\nan invocation API. Its security posture is the intersection of **three\nindependent resource objects** — the Model, the EndpointConfig, and the\nEndpoint — and the most common audit mistake is checking only one of them.\n\nA SageMaker real-time endpoint is not a single object. The security-relevant\nfields are split across a resource chain:\n\n- **Model** (`DescribeModel`) — holds `ExecutionRoleArn` (the IAM role the\n  container assumes), `VpcConfig` (network placement), `EnableNetworkIsolation`\n  (container egress control), and the container image spec.\n- **EndpointConfig** (`DescribeEndpointConfig`) — holds `KmsKeyId` (encryption\n  at rest), `ProductionVariants` (instance type + count), `DataCaptureConfig`\n  (request\u002Fresponse capture for monitoring), and\n  `EnableInterContainerTrafficEncryption`.\n- **Endpoint** (`DescribeEndpoint`) — holds `EndpointStatus` (InService \u002F\n  Updating \u002F Failed), `FailureReason`, and the resolved variant instance counts.\n- **MonitoringSchedule** (`ListMonitoringSchedules`) — optionally attached to\n  the endpoint for drift detection (data quality, model quality, bias,\n  explainability).\n\nAn audit that checks only the EndpointConfig misses the execution role and VPC\nplacement (both live on the Model). An audit that checks only the Model misses\nencryption and data capture (both live on the EndpointConfig). **Always resolve\nthe full chain.**\n\n## Quick reference — verdict thresholds\n\n| Condition | Verdict | Step |\n|---|---|---|\n| No `VpcConfig` on the Model (endpoint is internet-facing) | **PUBLIC_ENDPOINT** | 1 |\n| `VpcConfig` present but `SecurityGroupIds` allow `0.0.0.0\u002F0` inbound | **PUBLIC_ENDPOINT** | 1 |\n| No `KmsKeyId` on EndpointConfig | **NO_ENCRYPTION** | 2 |\n| Inference pipeline (multi-container) + `EnableInterContainerTrafficEncryption: false` | **NO_ENCRYPTION** | 2 |\n| Execution role with `Action: \"*\"` on `Resource: \"*\"` | **OVERPERMISSIVE_ROLE** | 3 |\n| Execution role with `sagemaker:*` \u002F `s3:*` \u002F `iam:*` on `\"*\"` | **OVERPERMISSIVE_ROLE** | 3 |\n| Execution role with `iam:PassRole` on `\"*\"` | **OVERPERMISSIVE_ROLE** | 3 |\n| No `DataCaptureConfig` (or `EnableCapture: false`) AND no `MonitoringSchedule` | **NO_MONITORING** | 4 |\n| `EnableCapture: true` but `InitialSamplingPercentage: 0` | **NO_MONITORING** | 4 |\n| `InitialInstanceCount \u003C 2` (single instance — no HA) | **CONFIG_GAP** | 5 |\n| All dimensions pass | **OK** | 6 |\n\n**Precedence (worst finding wins):**\nPUBLIC_ENDPOINT > NO_ENCRYPTION > OVERPERMISSIVE_ROLE > NO_MONITORING > CONFIG_GAP > OK\n\n## Pre-flight: endpoint metadata gate\n\nBefore evaluating the endpoint configuration, classify the endpoint state.\nSeveral endpoint attributes short-circuit the audit.\n\n| Attribute | Value | Effect on audit |\n|---|---|---|\n| `EndpointStatus` | `InService` | Proceed with full audit. |\n| `EndpointStatus` | `Updating` | Proceed, but note that config may be mid-change. Re-audit after update completes. |\n| `EndpointStatus` | `Failed` | Endpoint deployment failed. Note `FailureReason`. Still audit the config (the security findings may explain the failure). |\n| `EndpointStatus` | `Deleting` | Endpoint is being torn down. Audit for forensic record only. |\n| Endpoint type | Real-time | Full audit applies. |\n| Endpoint type | Async | Full audit + check `AsyncInferenceConfig.OutputConfig.S3OutputPath` is encrypted (KMS on the bucket). |\n| Endpoint type | Serverless | No instance count dimension (scale-to-zero). Skip Step 5 instance-count check. VPC config still applies. |\n\n**If the configuration bundle is incomplete** (missing Model, EndpointConfig,\nor Endpoint metadata), output:\n\n```text\nENDPOINT: \u003Cname>\nVERDICT: ERROR\nREASON: Cannot resolve the full Model -> EndpointConfig -> Endpoint resource chain — \u003Cmissing object>.\nREMEDIATION: Retrieve all three objects: aws sagemaker describe-model --model-name \u003Cname>, aws sagemaker describe-endpoint-config --endpoint-config-name \u003Cconfig>, aws sagemaker describe-endpoint --endpoint-name \u003Cname>.\n```\n\n## Process — Classification logic (apply in order, aggregate worst)\n\n### Step 0: Expert knowledge — non-obvious SageMaker behaviors\n\nNon-obvious SageMaker behaviors — VpcConfig lives on the Model, network isolation vs VpcConfig, inter-container encryption only for pipelines, TLS-by-default invocation, execution role on the Model, capture vs monitoring, zero sampling, instance counts, serverless\u002Fasync exceptions, monitor types, Clarify vs Model Monitor, async S3 output — live in [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad that reference before classifying an endpoint.\n\n### Step 1: VPC configuration — PUBLIC_ENDPOINT (highest severity)\n\nIf the Model has no `VpcConfig` (absent or empty), the endpoint is deployed on\nAWS-managed public infrastructure — **the invocation API is internet-facing**.\nAny client with the endpoint URL and valid AWS credentials can invoke the\nmodel. This is the most exposed posture.\n\n```text\nVERDICT: PUBLIC_ENDPOINT\nREASON: Model has no VpcConfig — endpoint is deployed on internet-facing\ninfrastructure. The InvokeEndpoint API is reachable from any network.\n```\n\n**Security-group refinement:** if `VpcConfig` IS present, inspect the\n`SecurityGroupIds`. A security group with an inbound rule allowing `0.0.0.0\u002F0`\non any port makes the endpoint effectively public within the VPC. Flag as\nPUBLIC_ENDPOINT.\n\n**EnableNetworkIsolation note:** a Model with no `VpcConfig` but\n`EnableNetworkIsolation: true` prevents the container from making outbound\ncalls, but the endpoint API is STILL internet-reachable for invocation. The\nPUBLIC_ENDPOINT verdict stands — network isolation restricts egress, not\ningress to the endpoint API.\n\n### Step 2: Encryption — NO_ENCRYPTION\n\nEvaluate two encryption dimensions:\n\n**2a. KMS encryption at rest.** Check `KmsKeyId` on the EndpointConfig. If\nabsent, the EBS volumes attached to the endpoint instances and the model\nartifact storage are unencrypted at rest. A compromised instance or snapshot\nexposes model weights and artifacts.\n\n- No `KmsKeyId` on EndpointConfig → **NO_ENCRYPTION**.\n\n**2b. Inter-container traffic encryption (inference pipelines only).** If the\nModel has multiple containers (a `SerialInferencePipeline` — check\n`Containers` array length > 1 or `PrimaryContainer` + the inference pipeline\nconfig), check `EnableInterContainerTrafficEncryption` on the EndpointConfig.\n\n- Multi-container endpoint + `EnableInterContainerTrafficEncryption: false`\n  → **NO_ENCRYPTION** (plaintext traffic between containers on the same host).\n- Single-container endpoint → this flag is a no-op. Do NOT flag.\n\n**2c. Async output S3 encryption (additive).** For async endpoints, if\n`AsyncInferenceConfig.OutputConfig.S3OutputPath` points to an S3 bucket\nwithout SSE-KMS encryption, append an additive NO_ENCRYPTION finding.\n\n### Step 3: Execution role blast radius — OVERPERMISSIVE_ROLE\n\nEvaluate the Model's `ExecutionRoleArn` trust policy (the identity-based\npolicy attached to the role). The execution role determines what AWS resources\nthe model container can access at runtime. Apply the same classification as\nthe iam-least-privilege-advisor:\n\n- `Action: \"*\"` on `Resource: \"*\"` → **OVERPERMISSIVE_ROLE** (admin-equivalent\n  — the container can do anything in the account).\n- Service wildcard on `\"*\"` (`sagemaker:*`, `s3:*`, `iam:*`, `kms:*`,\n  `secretsmanager:*`) → **OVERPERMISSIVE_ROLE**. The execution role only needs\n  narrow permissions (InvokeEndpoint, S3 GetObject on model artifacts, KMS\n  Decrypt on the endpoint key). Service wildcards are always over-scoped.\n- `iam:PassRole` on `\"*\"` → **OVERPERMISSIVE_ROLE** (privilege escalation —\n  the container can pass any role to any service).\n- `NotAction` \u002F `NotResource` → **OVERPERMISSIVE_ROLE** (inverse wildcards).\n- Named actions on specific ARNs → pass this step.\n\n**The SageMaker execution-role minimum-viable permission set:**\n`sagemaker:InvokeEndpoint` (self), `s3:GetObject` on the model artifact\nbucket\u002Fprefix, `kms:Decrypt` on the endpoint KMS key, and optionally\n`logs:CreateLogStream` \u002F `logs:PutLogEvents` on the CloudWatch log group.\nAnything beyond this scope should be justified by the model's runtime\nbehaviour (e.g., calling a downstream API, reading a feature store).\n\nIf no execution-role policy document is provided, note:\n\"Execution-role policy not provided — cannot evaluate IAM blast radius.\nRetrieve with `aws iam list-attached-role-policies --role-name \u003Cname>` and\n`aws iam get-role-policy --role-name \u003Cname> --policy-name \u003Cname>`.\"\n\n### Step 4: Monitoring coverage — NO_MONITORING\n\nEvaluate two monitoring dimensions:\n\n**4a. Data capture.** Check `DataCaptureConfig` on the EndpointConfig.\n\n- No `DataCaptureConfig` → data capture is disabled.\n- `EnableCapture: false` → data capture is disabled.\n- `EnableCapture: true` but `InitialSamplingPercentage: 0` → effectively\n  disabled (zero requests captured).\n\n**4b. Monitoring schedule.** Check for a `MonitoringSchedule` referencing the\nendpoint name.\n\n- No monitoring schedule → no drift detection.\n- Monitoring schedule with `Status: Stopped` → schedule exists but is inactive.\n\n**Classification:**\n- No data capture AND no monitoring schedule → **NO_MONITORING**.\n- Data capture enabled but no monitoring schedule → **NO_MONITORING** (capture\n  alone provides no alerts; the data sits idle without a schedule to analyse\n  it).\n- Monitoring schedule exists but data capture disabled → **NO_MONITORING**\n  (the schedule has no data to analyse — Model Monitor cannot function without\n  captured request\u002Fresponse payloads).\n- Both data capture (with sampling > 0) AND an active monitoring schedule →\n  pass this step.\n\n### Step 5: Instance count \u002F HA — CONFIG_GAP\n\nCheck `InitialInstanceCount` (or `CurrentInstanceCount` if available) on the\nendpoint's `ProductionVariants`.\n\n- `InitialInstanceCount \u003C 2` → **CONFIG_GAP** (single instance — no high\n  availability. If the instance fails, the endpoint goes down with no\n  fallback. Production endpoints should have at least 2 instances across\n  different Availability Zones).\n- Serverless endpoint → skip this step (scale-to-zero by design).\n- Async endpoint with `InitialInstanceCount: 1` → note as CONFIG_GAP but\n  acknowledge the cost-optimisation trade-off (async endpoints can scale to\n  zero; flag only if the endpoint serves latency-sensitive production\n  traffic).\n\n### Step 6: All dimensions pass — OK\n\nIf all of the following hold:\n- Model has `VpcConfig` with restrictive security groups.\n- EndpointConfig has `KmsKeyId`.\n- (Multi-container endpoints have `EnableInterContainerTrafficEncryption: true`).\n- Execution role has named actions on specific ARNs (no wildcards).\n- Data capture is enabled with sampling > 0 AND an active monitoring schedule\n  exists.\n- `InitialInstanceCount >= 2` (or serverless\u002Fasync with justification).\n\n→ **VERDICT: OK**\n\n### Step 7: Aggregation\n\nThe final verdict is the **worst finding** across all steps, where:\nPUBLIC_ENDPOINT > NO_ENCRYPTION > OVERPERMISSIVE_ROLE > NO_MONITORING > CONFIG_GAP > OK\n\nAll individual findings are listed in the FINDINGS section regardless of the\naggregate verdict.\n\n## Output format (per endpoint)\n\n```text\nENDPOINT: \u003Cendpoint-name>\nVERDICT: PUBLIC_ENDPOINT | NO_ENCRYPTION | OVERPERMISSIVE_ROLE | NO_MONITORING | CONFIG_GAP | OK\nREASON: \u003C1-2 sentences citing the worst finding and its step number>\nFINDINGS:\n  - [PUBLIC_ENDPOINT] \u003Cfinding description (Step N)>\n  - [NO_MONITORING] \u003Cfinding description (Step N)>\n  - [OK] \u003Cdimension that passed>\nREMEDIATION: \u003Cspecific CLI action per finding, or \"None required\" if OK>\n```\n\n### Worked example — public endpoint with no monitoring\n\n```text\nENDPOINT: public-endpoint-no-vpc\nVERDICT: PUBLIC_ENDPOINT\nREASON: Model has no VpcConfig — the endpoint is deployed on internet-facing\ninfrastructure (Step 1). Data capture is also disabled with no monitoring\nschedule (Step 4).\nFINDINGS:\n  - [PUBLIC_ENDPOINT] No VpcConfig on Model — InvokeEndpoint API reachable\n    from any network (Step 1)\n  - [NO_MONITORING] No DataCaptureConfig and no MonitoringSchedule — drift\n    undetectable (Step 4)\n  - [OK] KmsKeyId present on EndpointConfig (Step 2)\n  - [OK] Execution role scoped to named actions (Step 3)\nREMEDIATION:\n  1. Attach a VpcConfig to the Model:\n     aws sagemaker create-model --model-name \u003Cname> --containers \u003Cspec>\n       --vpc-config Subnets=\u003Csubnet-ids>,SecurityGroupIds=\u003Csg-ids>\n  2. Enable data capture on the EndpointConfig and create a monitoring\n     schedule (see Remediation guidance below).\n```\n\n## Anti-Patterns — NEVER\n\n- NEVER conclude an endpoint is \"private\" or \"VPC-attached\" by checking only\n  the EndpointConfig. `VpcConfig` lives on the **Model** — the EndpointConfig\n  has no VPC fields. Checking the wrong object is the most common false\n  negative in SageMaker endpoint audits.\n\n- NEVER flag `EnableInterContainerTrafficEncryption: false` on a\n  **single-container** endpoint. There is no inter-container traffic to\n  encrypt. The flag is a no-op. Flagging it is a false positive that erodes\n  trust.\n\n- NEVER classify an endpoint without `VpcConfig` but with\n  `EnableNetworkIsolation: true` as \"private.\" Network isolation blocks\n  container **egress**; it does not prevent inbound invocation. The\n  InvokeEndpoint API is still internet-reachable. The verdict is\n  PUBLIC_ENDPOINT.\n\n- NEVER assume the execution role is on the EndpointConfig or Endpoint. It is\n  on the **Model** (`ExecutionRoleArn`). Scanning the wrong object yields no\n  role and a silent pass on the IAM dimension.\n\n- NEVER treat `EnableCapture: true` as \"monitoring enabled.\" Data capture\n  stores request\u002Fresponse payloads to S3 — it provides no alerts. A\n  MonitoringSchedule is required to analyse the captured data and emit drift\n  findings. Both must be present.\n\n- NEVER accept `InitialSamplingPercentage: 0` as valid data capture. Zero\n  percent sampling captures zero requests. The operator may believe monitoring\n  is active when it is silently inert.\n\n- NEVER flag a serverless endpoint for `InitialInstanceCount` being absent or\n  below 2. Serverless endpoints use `ServerlessConfig` (memory size + max\n  concurrency), not instance count. The HA dimension does not apply.\n\n- NEVER assume `EndpointStatus: Failed` means the endpoint has no security\n  findings. A failed deployment often fails BECAUSE of a security\n  misconfiguration (e.g., execution role missing `sagemaker:CreateModel`,\n  VPC lacking S3 VPC endpoint for model artifacts). Audit the config — the\n  security finding may explain the failure.\n\n- NEVER recommend `sagemaker:*` as a remediation for the execution role. The\n  execution role needs `sagemaker:InvokeEndpoint` at most — `sagemaker:*`\n  grants model\u002Fendpoint creation and deletion, which the container never\n  needs at runtime. This reintroduces the exact blast radius being constrained.\n\n- NEVER overlook the async OutputConfig S3 path for async endpoints. Async\n  inference outputs are written to the configured S3 path. If the bucket\n  lacks SSE-KMS, outputs are unencrypted at rest — a silent data-exposure gap.\n\n- NEVER treat SageMaker Clarify enablement as a substitute for Model Monitor.\n  Clarify runs at build time (pre-training bias, explainability baselines).\n  Model Monitor runs at inference time (runtime drift detection). They are\n  different services with different triggers.\n\n- NEVER conflate `InitialInstanceCount` (EndpointConfig) with\n  `CurrentInstanceCount` (DescribeEndpoint runtime). Auto-scaling can change\n  the count after deployment. For HA auditing, prefer the runtime value when\n  available.\n\n## Pre-flight safety checks (run before any remediation CLI)\n\nThe full pre-flight safety checklist — confirm gate, Model\u002FEndpointConfig immutability, snapshot for rollback, caller permission check, VPC endpoint prerequisites — lives in [references\u002Fdiagnostic-commands.md](references\u002Fdiagnostic-commands.md).\nRun these before any remediation CLI.\n\n## Remediation guidance\n\n### For PUBLIC_ENDPOINT — no VpcConfig\n\n1. Create a new Model version with VpcConfig:\n   ```bash\n   aws sagemaker create-model --model-name \u003Cname>-vpc \\\n     --execution-role-arn \u003Crole-arn> --primary-container \u003Ccontainer-spec> \\\n     --vpc-config Subnets=subnet-abc,subnet-def,SecurityGroupIds=sg-xyz \\\n     --enable-network-isolation\n   ```\n2. Create a new EndpointConfig referencing the new model.\n3. Update the endpoint (blue\u002Fgreen):\n   ```bash\n   aws sagemaker update-endpoint --endpoint-name \u003Cname> \\\n     --endpoint-config-name \u003Cnew-config>\n   ```\n4. Verify the endpoint is now VPC-attached:\n   ```bash\n   aws sagemaker describe-model --model-name \u003Cname>-vpc --query 'VpcConfig'\n   ```\n\n### For NO_ENCRYPTION — no KMS key\n\n1. Create a new EndpointConfig with a customer-managed KMS key:\n   ```bash\n   aws sagemaker create-endpoint-config --endpoint-config-name \u003Cname>-enc \\\n     --production-variants \u003Cvariants> \\\n     --kms-key-id arn:aws:kms:us-east-1:\u003Cacct>:key\u002F\u003Ckey-id>\n   ```\n2. For inference pipelines, add\n   `--enable-inter-container-traffic-encryption`.\n3. Update the endpoint to the new config.\n4. For async endpoints, ensure the OutputConfig S3 bucket has SSE-KMS enabled.\n\n### For OVERPERMISSIVE_ROLE — wildcard execution role\n\n1. Replace wildcard actions with the minimum-viable permission set:\n   ```json\n   {\n     \"Version\": \"2012-10-17\",\n     \"Statement\": [{\n       \"Effect\": \"Allow\",\n       \"Action\": [\n         \"sagemaker:InvokeEndpoint\",\n         \"s3:GetObject\"\n       ],\n       \"Resource\": [\n         \"arn:aws:sagemaker:us-east-1:\u003Cacct>:endpoint\u002F\u003Cname>\",\n         \"arn:aws:s3:::\u003Cmodel-bucket>\u002F\u003Cprefix>\u002F*\"\n       ]\n     }]\n   }\n   ```\n2. Add KMS decrypt scoped to the endpoint key if the endpoint uses SSE-KMS.\n3. Derive the exact action set from CloudTrail (90-day window of\n   `userIdentity.arn` matching the execution role).\n\n### For NO_MONITORING — no data capture or schedule\n\n1. Create a new EndpointConfig with DataCaptureConfig:\n   ```bash\n   aws sagemaker create-endpoint-config --endpoint-config-name \u003Cname>-cap \\\n     --production-variants \u003Cvariants> \\\n     --data-capture-config EnableCapture=true,InitialSamplingPercentage=100,DestinationS3Uri=s3:\u002F\u002F\u003Cbucket>\u002Fcapture\u002F,CaptureOptions=[{CaptureMode=Input},{CaptureMode=Output}]\n   ```\n2. Update the endpoint to the new config.\n3. Create a monitoring schedule:\n   ```bash\n   aws sagemaker create-monitoring-schedule \\\n     --monitoring-schedule-name \u003Cname>-dq-monitor \\\n     --endpoint-name \u003Cname> \\\n     --monitoring-type DataQuality \\\n     --monitoring-job-definition \u003Cjob-def>\n   ```\n\n### For CONFIG_GAP — single instance\n\n1. Update the endpoint weights and capacities:\n   ```bash\n   aws sagemaker update-endpoint-weights-and-capacities \\\n     --endpoint-name \u003Cname> \\\n     --desired-weights-and-capacities variantName=\u003Cvariant>,DesiredInstanceCount=2\n   ```\n2. For async endpoints with justified scale-to-zero, document the\n   cost-optimisation trade-off and suppress the finding.\n\n### For OK\n\n1. No remediation required.\n2. Recommend verifying the monitoring schedule is producing results:\n   `aws sagemaker describe-monitoring-schedule --monitoring-schedule-name \u003Cname>`.\n3. Recommend reviewing captured data quarterly to confirm sampling is\n  representative.\n\n## Recent AWS features (2024-2026)\n\nRecent AWS features 2024-2026 — Inference Components, zero-downtime updates, HyperPod, Unified Studio, EKS-based inference — live in [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad when auditing endpoints built on these features.\n\n## References (load on demand)\n\n- [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md) — Step-0 non-obvious SageMaker behaviors and recent AWS features (2024-2026)\n- [references\u002Fdiagnostic-commands.md](references\u002Fdiagnostic-commands.md) — pre-flight safety checks and rollback commands before any remediation\n\n## Domain\n\nAWS CloudOps \u002F SageMaker AI\u002FML Security & Production Readiness.\n\n## AWS documentation\n\n- **Amazon SageMaker Developer Guide** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fsagemaker\u002Flatest\u002Fdg\u002Fwhatis.html\n- **Security** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fsagemaker\u002Flatest\u002Fdg\u002Fsecurity.html\n- **SageMaker API Reference** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fsagemaker\u002Flatest\u002FAPIReference\u002F\n- **AWS CLI reference (sagemaker)** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fcli\u002Flatest\u002Freference\u002Fsagemaker\u002F\n- **Inference Components** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fsagemaker\u002Flatest\u002Fdg\u002Finference-components.html\n","# SageMaker Endpoint Auditor\n\n## Mindset\n\n**One-line takeaway:** a SageMaker endpoint is a running model exposed behind\nan invocation API. Its security posture is the intersection of **three\nindependent resource objects** — the Model, the EndpointConfig, and the\nEndpoint — and the most common audit mistake is checking only one of them.\n\nA SageMaker real-time endpoint is not a single object. The security-relevant\nfields are split across a resource chain:\n\n- **Model** (`DescribeModel`) — holds `ExecutionRoleArn` (the IAM role the\n  container assumes), `VpcConfig` (network placement), `EnableNetworkIsolation`\n  (container egress control), and the container image spec.\n- **EndpointConfig** (`DescribeEndpointConfig`) — holds `KmsKeyId` (encryption\n  at rest), `ProductionVariants` (instance type + count), `DataCaptureConfig`\n  (request\u002Fresponse capture for monitoring), and\n  `EnableInterContainerTrafficEncryption`.\n- **Endpoint** (`DescribeEndpoint`) — holds `EndpointStatus` (InService \u002F\n  Updating \u002F Failed), `FailureReason`, and the resolved variant instance counts.\n- **MonitoringSchedule** (`ListMonitoringSchedules`) — optionally attached to\n  the endpoint for drift detection (data quality, model quality, bias,\n  explainability).\n\nAn audit that checks only the EndpointConfig misses the execution role and VPC\nplacement (both live on the Model). An audit that checks only the Model misses\nencryption and data capture (both live on the EndpointConfig). **Always resolve\nthe full chain.**\n\n## Quick reference — verdict thresholds\n\n| Condition | Verdict | Step |\n|---|---|---|\n| No `VpcConfig` on the Model (endpoint is internet-facing) | **PUBLIC_ENDPOINT** | 1 |\n| `VpcConfig` present but `SecurityGroupIds` allow `0.0.0.0\u002F0` inbound | **PUBLIC_ENDPOINT** | 1 |\n| No `KmsKeyId` on EndpointConfig | **NO_ENCRYPTION** | 2 |\n| Inference pipeline (multi-container) + `EnableInterContainerTrafficEncryption: false` | **NO_ENCRYPTION** | 2 |\n| Execution role with `Action: \"*\"` on `Resource: \"*\"` | **OVERPERMISSIVE_ROLE** | 3 |\n| Execution role with `sagemaker:*` \u002F `s3:*` \u002F `iam:*` on `\"*\"` | **OVERPERMISSIVE_ROLE** | 3 |\n| Execution role with `iam:PassRole` on `\"*\"` | **OVERPERMISSIVE_ROLE** | 3 |\n| No `DataCaptureConfig` (or `EnableCapture: false`) AND no `MonitoringSchedule` | **NO_MONITORING** | 4 |\n| `EnableCapture: true` but `InitialSamplingPercentage: 0` | **NO_MONITORING** | 4 |\n| `InitialInstanceCount \u003C 2` (single instance — no HA) | **CONFIG_GAP** | 5 |\n| All dimensions pass | **OK** | 6 |\n\n**Precedence (worst finding wins):**\nPUBLIC_ENDPOINT > NO_ENCRYPTION > OVERPERMISSIVE_ROLE > NO_MONITORING > CONFIG_GAP > OK\n\n## Pre-flight: endpoint metadata gate\n\nBefore evaluating the endpoint configuration, classify the endpoint state.\nSeveral endpoint attributes short-circuit the audit.\n\n| Attribute | Value | Effect on audit |\n|---|---|---|\n| `EndpointStatus` | `InService` | Proceed with full audit. |\n| `EndpointStatus` | `Updating` | Proceed, but note that config may be mid-change. Re-audit after update completes. |\n| `EndpointStatus` | `Failed` | Endpoint deployment failed. Note `FailureReason`. Still audit the config (the security findings may explain the failure). |\n| `EndpointStatus` | `Deleting` | Endpoint is being torn down. Audit for forensic record only. |\n| Endpoint type | Real-time | Full audit applies. |\n| Endpoint type | Async | Full audit + check `AsyncInferenceConfig.OutputConfig.S3OutputPath` is encrypted (KMS on the bucket). |\n| Endpoint type | Serverless | No instance count dimension (scale-to-zero). Skip Step 5 instance-count check. VPC config still applies. |\n\n**If the configuration bundle is incomplete** (missing Model, EndpointConfig,\nor Endpoint metadata), output:\n\n```text\nENDPOINT: \u003Cname>\nVERDICT: ERROR\nREASON: Cannot resolve the full Model -> EndpointConfig -> Endpoint resource chain — \u003Cmissing object>.\nREMEDIATION: Retrieve all three objects: aws sagemaker describe-model --model-name \u003Cname>, aws sagemaker describe-endpoint-config --endpoint-config-name \u003Cconfig>, aws sagemaker describe-endpoint --endpoint-name \u003Cname>.\n```\n\n## Process — Classification logic (apply in order, aggregate worst)\n\n### Step 0: Expert knowledge — non-obvious SageMaker behaviors\n\nNon-obvious SageMaker behaviors — VpcConfig lives on the Model, network isolation vs VpcConfig, inter-container encryption only for pipelines, TLS-by-default invocation, execution role on the Model, capture vs monitoring, zero sampling, instance counts, serverless\u002Fasync exceptions, monitor types, Clarify vs Model Monitor, async S3 output — live in [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad that reference before classifying an endpoint.\n\n### Step 1: VPC configuration — PUBLIC_ENDPOINT (highest severity)\n\nIf the Model has no `VpcConfig` (absent or empty), the endpoint is deployed on\nAWS-managed public infrastructure — **the invocation API is internet-facing**.\nAny client with the endpoint URL and valid AWS credentials can invoke the\nmodel. This is the most exposed posture.\n\n```text\nVERDICT: PUBLIC_ENDPOINT\nREASON: Model has no VpcConfig — endpoint is deployed on internet-facing\ninfrastructure. The InvokeEndpoint API is reachable from any network.\n```\n\n**Security-group refinement:** if `VpcConfig` IS present, inspect the\n`SecurityGroupIds`. A security group with an inbound rule allowing `0.0.0.0\u002F0`\non any port makes the endpoint effectively public within the VPC. Flag as\nPUBLIC_ENDPOINT.\n\n**EnableNetworkIsolation note:** a Model with no `VpcConfig` but\n`EnableNetworkIsolation: true` prevents the container from making outbound\ncalls, but the endpoint API is STILL internet-reachable for invocation. The\nPUBLIC_ENDPOINT verdict stands — network isolation restricts egress, not\ningress to the endpoint API.\n\n### Step 2: Encryption — NO_ENCRYPTION\n\nEvaluate two encryption dimensions:\n\n**2a. KMS encryption at rest.** Check `KmsKeyId` on the EndpointConfig. If\nabsent, the EBS volumes attached to the endpoint instances and the model\nartifact storage are unencrypted at rest. A compromised instance or snapshot\nexposes model weights and artifacts.\n\n- No `KmsKeyId` on EndpointConfig → **NO_ENCRYPTION**.\n\n**2b. Inter-container traffic encryption (inference pipelines only).** If the\nModel has multiple containers (a `SerialInferencePipeline` — check\n`Containers` array length > 1 or `PrimaryContainer` + the inference pipeline\nconfig), check `EnableInterContainerTrafficEncryption` on the EndpointConfig.\n\n- Multi-container endpoint + `EnableInterContainerTrafficEncryption: false`\n  → **NO_ENCRYPTION** (plaintext traffic between containers on the same host).\n- Single-container endpoint → this flag is a no-op. Do NOT flag.\n\n**2c. Async output S3 encryption (additive).** For async endpoints, if\n`AsyncInferenceConfig.OutputConfig.S3OutputPath` points to an S3 bucket\nwithout SSE-KMS encryption, append an additive NO_ENCRYPTION finding.\n\n### Step 3: Execution role blast radius — OVERPERMISSIVE_ROLE\n\nEvaluate the Model's `ExecutionRoleArn` trust policy (the identity-based\npolicy attached to the role). The execution role determines what AWS resources\nthe model container can access at runtime. Apply the same classification as\nthe iam-least-privilege-advisor:\n\n- `Action: \"*\"` on `Resource: \"*\"` → **OVERPERMISSIVE_ROLE** (admin-equivalent\n  — the container can do anything in the account).\n- Service wildcard on `\"*\"` (`sagemaker:*`, `s3:*`, `iam:*`, `kms:*`,\n  `secretsmanager:*`) → **OVERPERMISSIVE_ROLE**. The execution role only needs\n  narrow permissions (InvokeEndpoint, S3 GetObject on model artifacts, KMS\n  Decrypt on the endpoint key). Service wildcards are always over-scoped.\n- `iam:PassRole` on `\"*\"` → **OVERPERMISSIVE_ROLE** (privilege escalation —\n  the container can pass any role to any service).\n- `NotAction` \u002F `NotResource` → **OVERPERMISSIVE_ROLE** (inverse wildcards).\n- Named actions on specific ARNs → pass this step.\n\n**The SageMaker execution-role minimum-viable permission set:**\n`sagemaker:InvokeEndpoint` (self), `s3:GetObject` on the model artifact\nbucket\u002Fprefix, `kms:Decrypt` on the endpoint KMS key, and optionally\n`logs:CreateLogStream` \u002F `logs:PutLogEvents` on the CloudWatch log group.\nAnything beyond this scope should be justified by the model's runtime\nbehaviour (e.g., calling a downstream API, reading a feature store).\n\nIf no execution-role policy document is provided, note:\n\"Execution-role policy not provided — cannot evaluate IAM blast radius.\nRetrieve with `aws iam list-attached-role-policies --role-name \u003Cname>` and\n`aws iam get-role-policy --role-name \u003Cname> --policy-name \u003Cname>`.\"\n\n### Step 4: Monitoring coverage — NO_MONITORING\n\nEvaluate two monitoring dimensions:\n\n**4a. Data capture.** Check `DataCaptureConfig` on the EndpointConfig.\n\n- No `DataCaptureConfig` → data capture is disabled.\n- `EnableCapture: false` → data capture is disabled.\n- `EnableCapture: true` but `InitialSamplingPercentage: 0` → effectively\n  disabled (zero requests captured).\n\n**4b. Monitoring schedule.** Check for a `MonitoringSchedule` referencing the\nendpoint name.\n\n- No monitoring schedule → no drift detection.\n- Monitoring schedule with `Status: Stopped` → schedule exists but is inactive.\n\n**Classification:**\n- No data capture AND no monitoring schedule → **NO_MONITORING**.\n- Data capture enabled but no monitoring schedule → **NO_MONITORING** (capture\n  alone provides no alerts; the data sits idle without a schedule to analyse\n  it).\n- Monitoring schedule exists but data capture disabled → **NO_MONITORING**\n  (the schedule has no data to analyse — Model Monitor cannot function without\n  captured request\u002Fresponse payloads).\n- Both data capture (with sampling > 0) AND an active monitoring schedule →\n  pass this step.\n\n### Step 5: Instance count \u002F HA — CONFIG_GAP\n\nCheck `InitialInstanceCount` (or `CurrentInstanceCount` if available) on the\nendpoint's `ProductionVariants`.\n\n- `InitialInstanceCount \u003C 2` → **CONFIG_GAP** (single instance — no high\n  availability. If the instance fails, the endpoint goes down with no\n  fallback. Production endpoints should have at least 2 instances across\n  different Availability Zones).\n- Serverless endpoint → skip this step (scale-to-zero by design).\n- Async endpoint with `InitialInstanceCount: 1` → note as CONFIG_GAP but\n  acknowledge the cost-optimisation trade-off (async endpoints can scale to\n  zero; flag only if the endpoint serves latency-sensitive production\n  traffic).\n\n### Step 6: All dimensions pass — OK\n\nIf all of the following hold:\n- Model has `VpcConfig` with restrictive security groups.\n- EndpointConfig has `KmsKeyId`.\n- (Multi-container endpoints have `EnableInterContainerTrafficEncryption: true`).\n- Execution role has named actions on specific ARNs (no wildcards).\n- Data capture is enabled with sampling > 0 AND an active monitoring schedule\n  exists.\n- `InitialInstanceCount >= 2` (or serverless\u002Fasync with justification).\n\n→ **VERDICT: OK**\n\n### Step 7: Aggregation\n\nThe final verdict is the **worst finding** across all steps, where:\nPUBLIC_ENDPOINT > NO_ENCRYPTION > OVERPERMISSIVE_ROLE > NO_MONITORING > CONFIG_GAP > OK\n\nAll individual findings are listed in the FINDINGS section regardless of the\naggregate verdict.\n\n## Output format (per endpoint)\n\n```text\nENDPOINT: \u003Cendpoint-name>\nVERDICT: PUBLIC_ENDPOINT | NO_ENCRYPTION | OVERPERMISSIVE_ROLE | NO_MONITORING | CONFIG_GAP | OK\nREASON: \u003C1-2 sentences citing the worst finding and its step number>\nFINDINGS:\n  - [PUBLIC_ENDPOINT] \u003Cfinding description (Step N)>\n  - [NO_MONITORING] \u003Cfinding description (Step N)>\n  - [OK] \u003Cdimension that passed>\nREMEDIATION: \u003Cspecific CLI action per finding, or \"None required\" if OK>\n```\n\n### Worked example — public endpoint with no monitoring\n\n```text\nENDPOINT: public-endpoint-no-vpc\nVERDICT: PUBLIC_ENDPOINT\nREASON: Model has no VpcConfig — the endpoint is deployed on internet-facing\ninfrastructure (Step 1). Data capture is also disabled with no monitoring\nschedule (Step 4).\nFINDINGS:\n  - [PUBLIC_ENDPOINT] No VpcConfig on Model — InvokeEndpoint API reachable\n    from any network (Step 1)\n  - [NO_MONITORING] No DataCaptureConfig and no MonitoringSchedule — drift\n    undetectable (Step 4)\n  - [OK] KmsKeyId present on EndpointConfig (Step 2)\n  - [OK] Execution role scoped to named actions (Step 3)\nREMEDIATION:\n  1. Attach a VpcConfig to the Model:\n     aws sagemaker create-model --model-name \u003Cname> --containers \u003Cspec>\n       --vpc-config Subnets=\u003Csubnet-ids>,SecurityGroupIds=\u003Csg-ids>\n  2. Enable data capture on the EndpointConfig and create a monitoring\n     schedule (see Remediation guidance below).\n```\n\n## Anti-Patterns — NEVER\n\n- NEVER conclude an endpoint is \"private\" or \"VPC-attached\" by checking only\n  the EndpointConfig. `VpcConfig` lives on the **Model** — the EndpointConfig\n  has no VPC fields. Checking the wrong object is the most common false\n  negative in SageMaker endpoint audits.\n\n- NEVER flag `EnableInterContainerTrafficEncryption: false` on a\n  **single-container** endpoint. There is no inter-container traffic to\n  encrypt. The flag is a no-op. Flagging it is a false positive that erodes\n  trust.\n\n- NEVER classify an endpoint without `VpcConfig` but with\n  `EnableNetworkIsolation: true` as \"private.\" Network isolation blocks\n  container **egress**; it does not prevent inbound invocation. The\n  InvokeEndpoint API is still internet-reachable. The verdict is\n  PUBLIC_ENDPOINT.\n\n- NEVER assume the execution role is on the EndpointConfig or Endpoint. It is\n  on the **Model** (`ExecutionRoleArn`). Scanning the wrong object yields no\n  role and a silent pass on the IAM dimension.\n\n- NEVER treat `EnableCapture: true` as \"monitoring enabled.\" Data capture\n  stores request\u002Fresponse payloads to S3 — it provides no alerts. A\n  MonitoringSchedule is required to analyse the captured data and emit drift\n  findings. Both must be present.\n\n- NEVER accept `InitialSamplingPercentage: 0` as valid data capture. Zero\n  percent sampling captures zero requests. The operator may believe monitoring\n  is active when it is silently inert.\n\n- NEVER flag a serverless endpoint for `InitialInstanceCount` being absent or\n  below 2. Serverless endpoints use `ServerlessConfig` (memory size + max\n  concurrency), not instance count. The HA dimension does not apply.\n\n- NEVER assume `EndpointStatus: Failed` means the endpoint has no security\n  findings. A failed deployment often fails BECAUSE of a security\n  misconfiguration (e.g., execution role missing `sagemaker:CreateModel`,\n  VPC lacking S3 VPC endpoint for model artifacts). Audit the config — the\n  security finding may explain the failure.\n\n- NEVER recommend `sagemaker:*` as a remediation for the execution role. The\n  execution role needs `sagemaker:InvokeEndpoint` at most — `sagemaker:*`\n  grants model\u002Fendpoint creation and deletion, which the container never\n  needs at runtime. This reintroduces the exact blast radius being constrained.\n\n- NEVER overlook the async OutputConfig S3 path for async endpoints. Async\n  inference outputs are written to the configured S3 path. If the bucket\n  lacks SSE-KMS, outputs are unencrypted at rest — a silent data-exposure gap.\n\n- NEVER treat SageMaker Clarify enablement as a substitute for Model Monitor.\n  Clarify runs at build time (pre-training bias, explainability baselines).\n  Model Monitor runs at inference time (runtime drift detection). They are\n  different services with different triggers.\n\n- NEVER conflate `InitialInstanceCount` (EndpointConfig) with\n  `CurrentInstanceCount` (DescribeEndpoint runtime). Auto-scaling can change\n  the count after deployment. For HA auditing, prefer the runtime value when\n  available.\n\n## Pre-flight safety checks (run before any remediation CLI)\n\nThe full pre-flight safety checklist — confirm gate, Model\u002FEndpointConfig immutability, snapshot for rollback, caller permission check, VPC endpoint prerequisites — lives in [references\u002Fdiagnostic-commands.md](references\u002Fdiagnostic-commands.md).\nRun these before any remediation CLI.\n\n## Remediation guidance\n\n### For PUBLIC_ENDPOINT — no VpcConfig\n\n1. Create a new Model version with VpcConfig:\n   ```bash\n   aws sagemaker create-model --model-name \u003Cname>-vpc \\\n     --execution-role-arn \u003Crole-arn> --primary-container \u003Ccontainer-spec> \\\n     --vpc-config Subnets=subnet-abc,subnet-def,SecurityGroupIds=sg-xyz \\\n     --enable-network-isolation\n   ```\n2. Create a new EndpointConfig referencing the new model.\n3. Update the endpoint (blue\u002Fgreen):\n   ```bash\n   aws sagemaker update-endpoint --endpoint-name \u003Cname> \\\n     --endpoint-config-name \u003Cnew-config>\n   ```\n4. Verify the endpoint is now VPC-attached:\n   ```bash\n   aws sagemaker describe-model --model-name \u003Cname>-vpc --query 'VpcConfig'\n   ```\n\n### For NO_ENCRYPTION — no KMS key\n\n1. Create a new EndpointConfig with a customer-managed KMS key:\n   ```bash\n   aws sagemaker create-endpoint-config --endpoint-config-name \u003Cname>-enc \\\n     --production-variants \u003Cvariants> \\\n     --kms-key-id arn:aws:kms:us-east-1:\u003Cacct>:key\u002F\u003Ckey-id>\n   ```\n2. For inference pipelines, add\n   `--enable-inter-container-traffic-encryption`.\n3. Update the endpoint to the new config.\n4. For async endpoints, ensure the OutputConfig S3 bucket has SSE-KMS enabled.\n\n### For OVERPERMISSIVE_ROLE — wildcard execution role\n\n1. Replace wildcard actions with the minimum-viable permission set:\n   ```json\n   {\n     \"Version\": \"2012-10-17\",\n     \"Statement\": [{\n       \"Effect\": \"Allow\",\n       \"Action\": [\n         \"sagemaker:InvokeEndpoint\",\n         \"s3:GetObject\"\n       ],\n       \"Resource\": [\n         \"arn:aws:sagemaker:us-east-1:\u003Cacct>:endpoint\u002F\u003Cname>\",\n         \"arn:aws:s3:::\u003Cmodel-bucket>\u002F\u003Cprefix>\u002F*\"\n       ]\n     }]\n   }\n   ```\n2. Add KMS decrypt scoped to the endpoint key if the endpoint uses SSE-KMS.\n3. Derive the exact action set from CloudTrail (90-day window of\n   `userIdentity.arn` matching the execution role).\n\n### For NO_MONITORING — no data capture or schedule\n\n1. Create a new EndpointConfig with DataCaptureConfig:\n   ```bash\n   aws sagemaker create-endpoint-config --endpoint-config-name \u003Cname>-cap \\\n     --production-variants \u003Cvariants> \\\n     --data-capture-config EnableCapture=true,InitialSamplingPercentage=100,DestinationS3Uri=s3:\u002F\u002F\u003Cbucket>\u002Fcapture\u002F,CaptureOptions=[{CaptureMode=Input},{CaptureMode=Output}]\n   ```\n2. Update the endpoint to the new config.\n3. Create a monitoring schedule:\n   ```bash\n   aws sagemaker create-monitoring-schedule \\\n     --monitoring-schedule-name \u003Cname>-dq-monitor \\\n     --endpoint-name \u003Cname> \\\n     --monitoring-type DataQuality \\\n     --monitoring-job-definition \u003Cjob-def>\n   ```\n\n### For CONFIG_GAP — single instance\n\n1. Update the endpoint weights and capacities:\n   ```bash\n   aws sagemaker update-endpoint-weights-and-capacities \\\n     --endpoint-name \u003Cname> \\\n     --desired-weights-and-capacities variantName=\u003Cvariant>,DesiredInstanceCount=2\n   ```\n2. For async endpoints with justified scale-to-zero, document the\n   cost-optimisation trade-off and suppress the finding.\n\n### For OK\n\n1. No remediation required.\n2. Recommend verifying the monitoring schedule is producing results:\n   `aws sagemaker describe-monitoring-schedule --monitoring-schedule-name \u003Cname>`.\n3. Recommend reviewing captured data quarterly to confirm sampling is\n  representative.\n\n## Recent AWS features (2024-2026)\n\nRecent AWS features 2024-2026 — Inference Components, zero-downtime updates, HyperPod, Unified Studio, EKS-based inference — live in [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad when auditing endpoints built on these features.\n\n## References (load on demand)\n\n- [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md) — Step-0 non-obvious SageMaker behaviors and recent AWS features (2024-2026)\n- [references\u002Fdiagnostic-commands.md](references\u002Fdiagnostic-commands.md) — pre-flight safety checks and rollback commands before any remediation\n\n## Domain\n\nAWS CloudOps \u002F SageMaker AI\u002FML Security & Production Readiness.\n\n## AWS documentation\n\n- **Amazon SageMaker Developer Guide** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fsagemaker\u002Flatest\u002Fdg\u002Fwhatis.html\n- **Security** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fsagemaker\u002Flatest\u002Fdg\u002Fsecurity.html\n- **SageMaker API Reference** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fsagemaker\u002Flatest\u002FAPIReference\u002F\n- **AWS CLI reference (sagemaker)** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fcli\u002Flatest\u002Freference\u002Fsagemaker\u002F\n- **Inference Components** — https:\u002F\u002Fdocs.aws.amazon.com\u002Fsagemaker\u002Flatest\u002Fdg\u002Finference-components.html\n",{"name":5,"license":15,"metadata":24,"description":7,"compatibility":39},{"tags":25,"phase":26,"author":27,"domain":28,"family":29,"version":30,"keywords":31,"complexity":32,"entry_point":33,"when_to_use":34,"requires_llm":35,"verdict_shape":36,"invocation_schema":37,"supports_pipeline":35,"activation_triggers":38},"sagemaker, ai-ml, security, endpoint, encryption, vpc, monitoring, audit","2","Jacky Chan — AWS Community Builder","aws-cloudops","AI\u002FML","0.1.0","SageMaker, endpoint, KMS encryption, inter-container encryption, execution role, VPC config, data capture, model monitor, instance count, high availability, production readiness, endpoint audit, inference pipeline, network isolation, sagemaker:InvokeEndpoint, blast radius","high","false","Reviewing a SageMaker endpoint before production deployment, checking endpoint encryption (KMS + inter-container), auditing the execution role for wildcard permissions, validating VPC configuration (private vs internet-facing), verifying model monitoring coverage (data capture + monitoring schedule), or checking instance count for high availability.","true","NO_ENCRYPTION | OVERPERMISSIVE_ROLE | NO_MONITORING | PUBLIC_ENDPOINT | CONFIG_GAP | OK","Input: either (a) a SageMaker endpoint configuration bundle (Model + EndpointConfig + Endpoint metadata + optional MonitoringSchedule + optional execution-role policy document), OR (b) an endpoint name\u002FARN for live-account audit. Output: deterministic ENDPOINT\u002FVERDICT\u002FREASON\u002FFINDINGS\u002F REMEDIATION block per endpoint, where VERDICT is one of NO_ENCRYPTION, OVERPERMISSIVE_ROLE, NO_MONITORING, PUBLIC_ENDPOINT, CONFIG_GAP, OK.","audit this SageMaker endpoint, is my endpoint encrypted, check endpoint VPC config, execution role too broad, is model monitoring enabled, endpoint production ready, SageMaker endpoint public, inter-container encryption","Agent runtime that reads SKILL.md (Claude Code, Cursor, Windsurf, Codex, Gemini). No AWS CLI required for offline config-document classification. Live-account audits use aws sagemaker describe-endpoint, describe-endpoint-config, describe-model, and list-monitoring-schedules (AWS CLI v2, SSO or key-based credentials).","main","eeace51bd5dd1db2030729d83eeab988861553aa","skills\u002Fsagemaker-endpoint-auditor",0,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills","https:\u002F\u002Fgithub.com\u002Fjhkchan","CTO & Co-Founder @Beever-AI  & @votee. CAISP. AWS Community Builder. Organizer@GDGCloudHK. MongoDB Champion. Technopreneur + Infornographer.","2026-09-24 08:54:10",{"source":8,"skillName":5,"commands":49,"prompt":59,"zipAvailable":60},[50,53,56],{"tool":51,"command":52},"npx","npx skills add jhkchan\u002Faws-agent-skills --skill sagemaker-endpoint-auditor",{"tool":54,"command":55},"bunx","bunx skills add jhkchan\u002Faws-agent-skills --skill sagemaker-endpoint-auditor",{"tool":57,"command":58},"pnpm","pnpm dlx skills add jhkchan\u002Faws-agent-skills --skill sagemaker-endpoint-auditor","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002FSKILL.md",true,[62,68,72,76,80,84,88,92,96,100,104,108,112,116,120,124,128,132,136],{"path":63,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":66,"sourceHtmlUrl":67},"eval\u002Ftest-cases.yaml","REFERENCE","参考文档",16504,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Feval\u002Ftest-cases.yaml",{"path":69,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":70,"sourceHtmlUrl":71},"evals\u002Fbaselines\u002Fconfig-gap-single-instance.md",622,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fbaselines\u002Fconfig-gap-single-instance.md",{"path":73,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":74,"sourceHtmlUrl":75},"evals\u002Fbaselines\u002Fno-kms-encryption.md",590,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fbaselines\u002Fno-kms-encryption.md",{"path":77,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":78,"sourceHtmlUrl":79},"evals\u002Fbaselines\u002Fno-monitoring-no-capture.md",580,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fbaselines\u002Fno-monitoring-no-capture.md",{"path":81,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":82,"sourceHtmlUrl":83},"evals\u002Fbaselines\u002Fok-hardened-endpoint.md",689,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fbaselines\u002Fok-hardened-endpoint.md",{"path":85,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":86,"sourceHtmlUrl":87},"evals\u002Fbaselines\u002Foverpermissive-execution-role.md",650,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fbaselines\u002Foverpermissive-execution-role.md",{"path":89,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":90,"sourceHtmlUrl":91},"evals\u002Fbaselines\u002Fpublic-endpoint-no-vpc.md",800,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fbaselines\u002Fpublic-endpoint-no-vpc.md",{"path":93,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":94,"sourceHtmlUrl":95},"evals\u002Feval_config.yaml",1905,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Feval_config.yaml",{"path":97,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":98,"sourceHtmlUrl":99},"evals\u002Fevals.json",4889,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fevals.json",{"path":101,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":102,"sourceHtmlUrl":103},"evals\u002Fprompts\u002Fconfig-gap-single-instance.md",1967,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fprompts\u002Fconfig-gap-single-instance.md",{"path":105,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":106,"sourceHtmlUrl":107},"evals\u002Fprompts\u002Fno-kms-encryption.md",1942,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fprompts\u002Fno-kms-encryption.md",{"path":109,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":110,"sourceHtmlUrl":111},"evals\u002Fprompts\u002Fno-monitoring-no-capture.md",1606,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fprompts\u002Fno-monitoring-no-capture.md",{"path":113,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":114,"sourceHtmlUrl":115},"evals\u002Fprompts\u002Fok-hardened-endpoint.md",2064,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fprompts\u002Fok-hardened-endpoint.md",{"path":117,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":118,"sourceHtmlUrl":119},"evals\u002Fprompts\u002Foverpermissive-execution-role.md",1665,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fprompts\u002Foverpermissive-execution-role.md",{"path":121,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":122,"sourceHtmlUrl":123},"evals\u002Fprompts\u002Fpublic-endpoint-no-vpc.md",1861,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fevals\u002Fprompts\u002Fpublic-endpoint-no-vpc.md",{"path":125,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":126,"sourceHtmlUrl":127},"examples\u002Fend-to-end.md",6941,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fexamples\u002Fend-to-end.md",{"path":129,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":130,"sourceHtmlUrl":131},"references\u002Fadvanced-patterns.md",5943,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Freferences\u002Fadvanced-patterns.md",{"path":133,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":134,"sourceHtmlUrl":135},"references\u002Fdiagnostic-commands.md",1958,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Freferences\u002Fdiagnostic-commands.md",{"path":137,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":138,"sourceHtmlUrl":139},"skill.usf.yaml",4196,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsagemaker-endpoint-auditor\u002Fskill.usf.yaml",[141,148,155,162,169,176],{"skillNo":142,"name":143,"nameZh":6,"description":144,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":145,"categories":146,"keywords":147,"publishedAt":19},"01M7ADVD60X4PCAPKSE58P9NDC","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 \u002F method \u002F 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 \u002F 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 \u002F EKS \u002F 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.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fxray-tracing-deployer\u002FSKILL.md",[],[],{"skillNo":149,"name":150,"nameZh":6,"description":151,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":152,"categories":153,"keywords":154,"publishedAt":19},"01M7ADTVJ7DCBM9CB1FY51GH7J","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.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwellarchitected-workload-auditor\u002FSKILL.md",[],[],{"skillNo":156,"name":157,"nameZh":6,"description":158,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":159,"categories":160,"keywords":161,"publishedAt":19},"01M7ADT59G6FTTNZYHT2QRPNNP","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 \u002F 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.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwellarchitected-review-operator\u002FSKILL.md",[],[],{"skillNo":163,"name":164,"nameZh":6,"description":165,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":166,"categories":167,"keywords":168,"publishedAt":19},"01M7ADSH5JA9QHNPPHQCD7EVA7","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\u002FAPI Gateway\u002FCloudFront, or configuring CAPTCHA\u002FChallenge 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.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwafv2-web-acl-deployer\u002FSKILL.md",[],[],{"skillNo":170,"name":171,"nameZh":6,"description":172,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":173,"categories":174,"keywords":175,"publishedAt":19},"01M7ADRYFAT8R73FBNCHRC573X","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\u002Fbypass 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.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwafv2-web-acl-auditor\u002FSKILL.md",[],[],{"skillNo":177,"name":178,"nameZh":6,"description":179,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":180,"categories":181,"keywords":182,"publishedAt":19},"01M7ADR8W7G72JX4KFGN03QK8Q","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.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwaf-rule-deployer\u002FSKILL.md",[],[]]