| name | haipipe-end-deploy-sagemaker |
| description | AWS SageMaker deploy specialist for haipipe-end: wraps an Endpoint_Set into SageMaker model.tar.gz, deploys to a SageMaker endpoint, runs live smoke tests, monitors logs, tears down. Reads (never modifies) Endpoint_Sets from haipipe-end-endpointset. Read the SageMaker pitfalls in ../haipipe-end-develop-sagemaker/SKILL.md before any live action. Called by /haipipe-end when deploy target is sagemaker. |
| argument-hint | [function] [endpoint_set_or_id] [args...] |
| allowed-tools | Bash, Read, Write, Edit, Grep, Glob |
| metadata | {"version":"0.1.2","last_updated":"2026-07-08","summary":"AWS SageMaker deploy specialist for haipipe-end."} |
Skill: haipipe-end-deploy-sagemaker
AWS SageMaker deployment specialist.
Consumes an Endpoint_Set built by haipipe-end-endpointset, packages it for SageMaker, deploys, tests live, monitors, and cleans up.
Function axis: dashboard | deploy | test | monitor | teardown | review
Commands
/haipipe-end-deploy-sagemaker -> dashboard: SageMaker endpoints + cost
/haipipe-end-deploy-sagemaker dashboard -> same
/haipipe-end-deploy-sagemaker deploy <endpoint_set> -> push Endpoint_Set to SageMaker
/haipipe-end-deploy-sagemaker test <endpoint_id> -> hit live SageMaker endpoint
/haipipe-end-deploy-sagemaker monitor <endpoint_id> -> CloudWatch logs + invocation metrics
/haipipe-end-deploy-sagemaker teardown <endpoint_id> -> stop endpoint, optional model cleanup
/haipipe-end-deploy-sagemaker review <endpoint_id> -> audit deploy config (IAM, instance, scaling)
Dispatch Table
Invocation Ref file(s) Function block
-------------- ---------------------------------------- -----------------------------------
dashboard ../haipipe-end/ref/deploy-overview.md dashboard procedure (in this SKILL.md)
deploy ../haipipe-end/ref/deploy-overview.md +
../haipipe-end/ref/
0-overview.md deploy procedure
test ../haipipe-end/ref/deploy-overview.md test procedure
monitor ../haipipe-end/ref/deploy-overview.md monitor procedure
teardown ../haipipe-end/ref/deploy-overview.md teardown procedure
review ../haipipe-end/ref/deploy-overview.md review procedure
The deploy step reads the endpointset overview to know the Endpoint_Set layout it consumes.
Step-by-Step Protocol
Step 0: Read ../haipipe-end/ref/deploy-overview.md for SageMaker-specific conventions
(instance types, IAM roles, model.tar.gz layout, autoscaling).
Step 1: Parse args.
Function vocabulary above.
Required arg per function:
deploy: <endpoint_set_name> (path under 6-EndpointStore/)
test/monitor/teardown/review: <sagemaker_endpoint_id>
Step 2: Verify AWS context:
- AWS credentials available (env or aws configure)
- Region set
- IAM role for SageMaker execution exists or can be created
Step 3: Execute the function (see procedures below).
Step 4: Emit the structured tail:
status: ok | blocked | failed
summary: 2-3 sentences on the deploy / test / etc.
artifacts: [SageMaker endpoint id, ARN, model package path on S3]
next: suggested next command
Procedures (placeholder — fill from project's actual SageMaker conventions)
Deploy:
- Read Endpoint_Set at
_WorkSpace/6-EndpointStore/<endpoint_set>/.
(input contract, all deploy skills: canonical input = the folder; a .tar.gz twin is a wire form only)
- Build
model.tar.gz containing fn_endpoint/ + ModelInstance + an (logical bundle name; physically materialized as code/ + model/ in the set)
inference.py entry point conforming to SageMaker's contract.
- Upload to S3 (project-configured bucket).
- Register a SageMaker Model pointing at the S3 artifact + execution role.
- Create EndpointConfig (instance type, autoscaling).
- Create Endpoint and wait for
InService.
- Run a 1-payload smoke invocation; verify response.
- Record endpoint_id + ARN in the project's deploy log.
Test, Monitor, Teardown, Review:
See ../haipipe-end/ref/deploy-overview.md for the SageMaker-specific procedures and the
aws sagemaker / aws cloudwatch CLI invocations the project uses.
Target Scope
Owns:
model.tar.gz packaging conforming to SageMaker's inference container contract
- SageMaker endpoint config (instance type, count, autoscaling)
- IAM role assumptions
aws sagemaker CLI invocation
- CloudWatch log queries
- Live invocation smoke tests against the deployed endpoint
Does NOT own:
- Endpoint_Set content (read-only input from
/haipipe-end-endpointset)
- ModelInstance training (
/haipipe-nn)
If a deploy fails because of an Endpoint_Set issue, escalate to /haipipe-end-endpointset review rather than patching here.