CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

aws-ai-ml

Selects, deploys, and customizes AI models on Amazon SageMaker. Training or Processing jobs, fine-tuning (SFT/DPO/RLVR/RLAIF), model selection, evaluation, SageMaker or Bedrock deployment, inference optimization and endpoint diagnostics. Use when fine-tuning models on SageMaker, choosing a base model to customize or deploy from SageMaker JumpStart or Hub, SageMakerPublicHub, or the model catalog, checking data quality (duplicates, languages, PII), transforming or validating data (reformat, clean, filter, dedupe, split, merge, redact), generating synthetic data, labeling data with a model, human review or annotation and applying its corrections, evaluating model quality, deploying or optimizing endpoints, configuring IAM/S3, or managing SageMaker Managed MLflow. Use for endpoint health, latency, logs, metrics, errors. Covers Serverless Model Customization, Nova and OSS deployment paths, and PySDK v3. NOT for SageMaker Ground Truth labeling jobs, Feature Store, or general AWS infrastructure.

Quality

87%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Failed to scan

The risk profile of this skill

SKILL.md
Quality
Evals
Security

Quality

Content

75%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-engineered router: an intent-to-reference table with explicit precedence rules, a concrete attribution command, and a thoughtful MCP-vs-local file-resolution guardrail, all backed by a real and consistently organized bundle. Its main cost is token weight — oversized trigger enumerations and repeated validated-workflow boilerplate inflate the table — and reference nesting goes a level deeper than ideal.

Suggestions

Trim the routing table's trigger-phrase enumerations (dataset-quality, auto-labeling, dataset-transformation rows) to a handful of representative phrases and move the full lists into each reference's overview.md, saving several hundred tokens per load.

State the "only validated workflow — do not attempt from general knowledge" rule once in the Rules section instead of repeating it across three routing rows.

Link each routing row directly to its folder's overview.md file (e.g. references/model-selection/overview.md) so navigation resolves in one hop instead of stopping at the folder.

DimensionReasoningScore

Conciseness

No concept-explanation padding and crisp Rules, but routing cells are token-heavy: ~25 quoted trigger phrases in the dataset-quality row, equally long lists in auto-labeling and dataset-transformation rows, the "ONLY validated workflow… do not attempt from general knowledge" boilerplate repeated across three rows, and long parenthetical expansions in the planning row. Not 4: the trimming opportunity is more than minor — the trigger enumerations could move to each reference's overview.

3 / 5

Actionability

Concrete, executable dispatch throughout: exact reference folder paths, a copy-paste command ("export AWS_SDK_UA_APP_ID=AWSSkill-SageMaker"), explicit decision procedures ("ask one service-choice question. Do not select a service by default"), and a guardrail giving literal file parameters for both MCP and local load modes. Routing instructions are unambiguous and immediately executable.

5 / 5

Workflow Clarity

Sequencing and precedence are explicit — "Load this reference FIRST when the request matches multiple rows", "route to use-case-specification first", and conflict resolution between auto-labeling and human-in-the-loop — but there are no validation checkpoints at the router level (they presumably live in the references). Not 5: no explicit verify/recover steps for mis-routes or failed loads in the body itself.

4 / 5

Progressive Disclosure

A clean overview-only SKILL.md routing to 18 real, consistently structured reference folders (all verified to exist, each with overview.md plus scripts/references), with the guardrail explaining how to resolve files. Not 5: links target folders rather than files, and content nests a second level deep (e.g. references/model-selection/references/select-for-deployment.md), so navigation is not strictly one-level-deep.

4 / 5

Total

16

/

20

Passed

Description

96%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

An exceptional description: third-person concrete capabilities, exhaustive natural trigger phrases with synonyms, and explicit what/when/when-not boundaries. Its only weakness is that broad data-transformation verbs and Bedrock/IAM scope introduce minor conflict risk with neighboring skills in an AWS skill suite.

DimensionReasoningScore

Specificity

Lists multiple concrete actions spanning the full lifecycle — "Selects, deploys, and customizes AI models", "Training or Processing jobs, fine-tuning (SFT/DPO/RLVR/RLAIF)", "inference optimization and endpoint diagnostics" — with comprehensive coverage. Not 4: the gaps are not minor; the action list covers selection, training, evaluation, deployment, diagnostics, and data operations.

5 / 5

Completeness

Explicitly answers what ("Selects, deploys, and customizes AI models on Amazon SageMaker…"), when ("Use when fine-tuning models on SageMaker…", "Use for endpoint health, latency, logs, metrics, errors"), and when-not ("NOT for SageMaker Ground Truth labeling jobs, Feature Store, or general AWS infrastructure").

5 / 5

Trigger Term Quality

Comprehensive natural-language triggers with synonyms: "dedupe"/"duplicates", "reformat, clean, filter, dedupe, split, merge, redact", "generate synthetic data", "endpoint health, latency, logs, metrics, errors", "human review or annotation". Users would naturally say these phrases when they need this skill.

5 / 5

Distinctiveness Conflict Risk

Clear SageMaker-model-customization niche with explicit exclusions, but generic data-operation triggers ("clean, filter, dedupe, split, merge") and the Bedrock/IAM/S3 coverage create minor overlap risk with adjacent AWS-infrastructure or data-processing skills. Not 5: those broad data verbs would plausibly fire for non-SageMaker data requests.

4 / 5

Total

19

/

20

Passed

Validation

93%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

referenced_paths_exist

Referenced path issues: 2 deeper-than-1-level

Warning

Total

15

/

16

Passed

Repository

aws/agent-toolkit-for-aws

Path

skills/core-skills/aws-ai-ml/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.